Security Now 1090 transcript
Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.
Leo Laporte [00:00:00]:
It's time for Security Now, and this is a very special episode. You're gonna be glad you're here. We are at the Black Hat Conference in Las Vegas. Steve Gibson's here with questions from you, the audience. We're gonna primarily talk about AI and security, and we've got some very special guests joining us. Security Now is next.
Steve Gibson [00:00:24]:
Podcasts you love.
Leo Laporte [00:00:25]:
From people you trust. This is TWiT. This is Security Now with Steve Gibson, episode 1090, recorded Wednesday, August 5th, 2026. Black Hat. It's time for Security Now, the show where we cover your security, your privacy, and how things work online. Hello everybody. Yes, it's a little bit different, a little bit noisier because we— and Steve can do that, which he could never do until recently.
Steve Gibson [00:01:00]:
I could touch you.
Leo Laporte [00:01:01]:
It's the miracle of sitting next to each other.
Paul Thurrott [00:01:03]:
No touching. No touching.
Leo Laporte [00:01:04]:
We are at Black Hat, the big security conference in Las Vegas, Nevada. Special guests of ThreatLocker. Thank you, ThreatLocker, for flying us all here. And Steve is here, but Steve is not alone. Say hi, Steve.
Steve Gibson [00:01:19]:
Hi, Steve.
Leo Laporte [00:01:20]:
Hi, Steve. You have especially asked, because we did Windows Weekly earlier, that Paul and Richard stick around.
Paul Thurrott [00:01:26]:
Yeah.
Leo Laporte [00:01:27]:
And they're all going to be part of the larger Security Now Because it's going to be kind of a different format this week.
Steve Gibson [00:01:33]:
Yeah, I kind of— but I thought it would be fun since we're here at ThreatLocker to— and these guys are also both here. We said, hey, you're not leaving.
Paul Thurrott [00:01:41]:
Yeah.
Steve Gibson [00:01:42]:
After your podcast, we need you. We want to do sort of a roundtable format. I asked our listeners last week to send in some thoughts that they thought would be fun talking points and just stuff about security, but naturally focused on AI. Because I mean, this whole conference could be renamed Applied AI for Security.
Leo Laporte [00:02:04]:
It's true.
Steve Gibson [00:02:05]:
I mean, if you're not doing AI now—
Richard Campbell [00:02:07]:
It's on every sign.
Steve Gibson [00:02:07]:
Don't even show up.
Leo Laporte [00:02:08]:
Yeah, exactly.
Paul Thurrott [00:02:10]:
Yeah.
Steve Gibson [00:02:10]:
So yeah, I think we're going to have a lot of fun.
Leo Laporte [00:02:12]:
And I'll just point out that Steve has brought paper.
Steve Gibson [00:02:16]:
Yes, it's, it's, you know, ThreatLog. It's Black Hat.
Leo Laporte [00:02:20]:
No Wi-Fi here.
Paul Thurrott [00:02:22]:
No.
Steve Gibson [00:02:22]:
And the battery lasts a long time.
Richard Campbell [00:02:24]:
Yeah.
Steve Gibson [00:02:24]:
And as I think Richard said—
Richard Campbell [00:02:27]:
Resolution's excellent.
Steve Gibson [00:02:28]:
Good high resolution screen. Really good resolution. Now my eyes were also high resolution.
Leo Laporte [00:02:33]:
You guys are doing all right. You guys are doing all right. We will get to the show and I know you didn't bring a picture of the week.
Steve Gibson [00:02:41]:
I did not.
Leo Laporte [00:02:41]:
But I brought a video of the week.
Paul Thurrott [00:02:43]:
Oh.
Leo Laporte [00:02:43]:
We will get to that in just a little bit. You're watching Security Now, a special live presentation from Black Hat.
Paul Thurrott [00:02:51]:
I love that.
Leo Laporte [00:02:52]:
He's not gonna resist, is he?
Paul Thurrott [00:02:53]:
No.
Leo Laporte [00:02:54]:
We'll have more right after this.
Steve Gibson [00:02:56]:
Paul remembers my Grinch.
Paul Thurrott [00:02:58]:
Yes, yes, right, exactly.
Leo Laporte [00:03:01]:
Back to Las Vegas and Black Hat with the entire gang. This is actually going to be so much fun. Steve Gibson is here, but you have asked Paul and Richard to come as well. And so we have kind of the fullest cast version of Security Now we've ever done. The full cast.
Steve Gibson [00:03:17]:
The full monty.
Leo Laporte [00:03:18]:
Full monty. Now it is a different format. As we said earlier, you're going to do some questions from the audience. We're all going to talk about whatever They asked, and normally at this point in the show we would do a picture of the week, but you told me last night, oh, I didn't bring a picture of the week.
Steve Gibson [00:03:33]:
Right, but because—
Leo Laporte [00:03:34]:
We're here at Black Hat and I ran into somebody. I'm not going to say who. I ran into somebody who is actually coming for DEF CON. DEF CON is the conference that immediately follows Black Hat that is more of the hackery.
Steve Gibson [00:03:47]:
It's the exhausted conference.
Paul Thurrott [00:03:49]:
Why is this called Black Hat and not White Hat?
Leo Laporte [00:03:53]:
Yeah.
Paul Thurrott [00:03:54]:
Black hats are the bad guys.
Leo Laporte [00:03:55]:
There is a history to this. The 2 were started by the same guy way back when, and they were intended to be hacker conferences. This has become the business corporate conference. And you're right, it should be white hat. It's more like RSAC, the RSA conference, where it's just a bunch of businesses that do security. The next conference, which actually overlaps a little bit, starts I think tomorrow. DEF CON is really more about hacking stuff. They're the ones that have, you know, the special hacking areas and things like that.
Leo Laporte [00:04:24]:
They are also the ones with the wall of sheep. where people who've been compromised during the conference get their names on the wall, things like that.
Steve Gibson [00:04:31]:
And remember, technically, we're not supposed to have any hat colors. You can't really have white and black anymore. So it's hats of unknown.
Leo Laporte [00:04:39]:
Hats of unknown colors.
Richard Campbell [00:04:40]:
Okay.
Leo Laporte [00:04:40]:
So I did run into somebody who was here for the HackerCon, and I said, oh, tell me a little bit about what you did. And then, and then as we talked, I said, one of the things I'm looking forward to in Las Vegas, Zoox, which is an Amazon company, has introduced a new autonomous ride. It's kind of competing with Waymo, but have you seen the little Zoox vehicles? They don't have a front seat, they don't have a driver's wheel.
Richard Campbell [00:05:03]:
Right.
Leo Laporte [00:05:04]:
They're committed. It's what they call campfire seating. You're in a living room.
Paul Thurrott [00:05:07]:
Right, like a London cab seat.
Leo Laporte [00:05:09]:
Yeah, and well, there's 2, but they're facing each other. Yeah, like only one of you can see the impending doom. Well, no, because it goes in both directions.
Richard Campbell [00:05:17]:
There you go.
Leo Laporte [00:05:17]:
In fact, the headlights turn into taillights and the taillights turn into headlights. It can arbitrarily go either way.
Steve Gibson [00:05:23]:
That makes parking much easier.
Richard Campbell [00:05:26]:
Absolutely.
Leo Laporte [00:05:26]:
And they're really cute. I don't know if you've seen them yet, but I— before we leave, I really want to ride a Zoox.
Paul Thurrott [00:05:30]:
So they have them here, in other words?
Leo Laporte [00:05:32]:
Yeah.
Paul Thurrott [00:05:32]:
Oh, I'd have to go outside to see that, so no, I'm not seeing it.
Leo Laporte [00:05:35]:
It's 109 degrees right now.
Richard Campbell [00:05:37]:
It's gonna burst into flames out there.
Leo Laporte [00:05:39]:
They were testing— they've been testing for a few months in Vegas. Uh, just a couple of days ago, NHTSA, the National Highway Transportation and Safety Administration, said you can charge people money for it. So they are now officially—
Richard Campbell [00:05:50]:
They're a taxi.
Leo Laporte [00:05:51]:
A taxi.
Richard Campbell [00:05:51]:
Wow.
Leo Laporte [00:05:52]:
In town. So I— maybe after the show we'll all go out. But I mentioned that I really want to ride in the Zoox. He says, oh yeah, we hacked the Zoox. I said, what?
Paul Thurrott [00:06:01]:
Oh boy.
Leo Laporte [00:06:02]:
He said, yeah, no, this was— it's all been fixed.
Paul Thurrott [00:06:05]:
Okay.
Leo Laporte [00:06:05]:
But I have a little video of the, uh, the attack. I think this is it.
Richard Campbell [00:06:19]:
Can't.
Leo Laporte [00:06:19]:
So— I had your track. Can you start over?
Paul Thurrott [00:06:24]:
You just cut to him.
Steve Gibson [00:06:25]:
I can't, but he's walking.
Leo Laporte [00:06:27]:
Stop it.
Richard Campbell [00:06:28]:
Go back to the wide, I think.
Leo Laporte [00:06:30]:
Okay.
Paul Thurrott [00:06:32]:
That went great.
Leo Laporte [00:06:35]:
So he said, We had a little fun with the Zooks a short while ago. Turns out there was a— he said, don't tell anybody how we did this. Amazon has fixed it, but we managed to get all the Zooks to show up all at once.
Paul Thurrott [00:06:54]:
Oh, I love it. Wow.
Leo Laporte [00:06:57]:
With one call.
Paul Thurrott [00:06:57]:
It's a Zook storm.
Leo Laporte [00:06:59]:
It's a Zook storm. Every single one of these Zooks. I said, how many were there? He says, many. Many Zooks.
Paul Thurrott [00:07:08]:
So this is in front of a—
Leo Laporte [00:07:09]:
This is at the casino. It might even be our casino. I don't know. It's one of the— one of the— you know how they drive up?
Steve Gibson [00:07:14]:
Did they have an exposed USB port?
Leo Laporte [00:07:16]:
No, this was all done, um, over the air, shall we say.
Steve Gibson [00:07:22]:
Okay.
Leo Laporte [00:07:22]:
But this video goes on. Let me just tell you, there's— wow, all the Zooks in Las Vegas came to the same place. Came to the same place.
Richard Campbell [00:07:30]:
Converged.
Paul Thurrott [00:07:31]:
Yeah. Which I have to say, if you're out there at 2 o'clock in the morning when this happened, this probably scared the living daylights out of you.
Richard Campbell [00:07:36]:
Oh my God, the robots are taking over.
Paul Thurrott [00:07:38]:
Yep, it's finally happening.
Leo Laporte [00:07:39]:
I promised I wouldn't name any names or talk about how they did it, but I thought, can I just show this video? So that is video of the week.
Steve Gibson [00:07:45]:
Perfect for the Suits score.
Leo Laporte [00:07:47]:
So, uh, tell us, Steve, what's—
Steve Gibson [00:07:50]:
okay, so what I wanted to start with is, uh, to sort of introduce each of us relative to our current framing of AI.
Paul Thurrott [00:08:00]:
Uh, um, this is only a 2-hour show. I don't know if we have any time.
Leo Laporte [00:08:04]:
I have some stories to tell.
Steve Gibson [00:08:06]:
No, no demos, Leo. We don't— we don't have time for demos.
Leo Laporte [00:08:09]:
Oh, rats.
Steve Gibson [00:08:10]:
Everybody knows That, uh, that I was— I'm a little bit of a Luddite or a little slow adopter. Paul loves the fact that I'm still coding in—
Paul Thurrott [00:08:19]:
I love it so much.
Steve Gibson [00:08:20]:
Assembly language.
Leo Laporte [00:08:21]:
Are you still using Windows 7?
Steve Gibson [00:08:23]:
I do have— I have not yet fully retired my Windows 7 machine.
Paul Thurrott [00:08:26]:
Have you learned ARM assembly language yet?
Steve Gibson [00:08:28]:
But no, and I don't think I will. Oh no, because RISC is not fun to program. Yeah, you know, a CISC chip, that's fun.
Leo Laporte [00:08:35]:
Yeah, he likes segmented memory, friends.
Richard Campbell [00:08:39]:
He likes Big instructions.
Paul Thurrott [00:08:40]:
That's what he's saying.
Leo Laporte [00:08:41]:
Give me the big instructions.
Steve Gibson [00:08:43]:
So I, I am yet to have any AI write any code for me.
Richard Campbell [00:08:48]:
Okay.
Leo Laporte [00:08:48]:
Although, wait a minute, are you big-endian or little-endian?
Steve Gibson [00:08:51]:
I'm, uh, little-endian.
Leo Laporte [00:08:53]:
Okay.
Steve Gibson [00:08:54]:
I'm in—
Paul Thurrott [00:08:54]:
I gotta think about it because in this world there is only that.
Steve Gibson [00:08:56]:
I was gonna say, wait, why have 2?
Paul Thurrott [00:08:58]:
Like, what do you mean? Why would there be both? Yeah.
Leo Laporte [00:09:00]:
Okay. That was a very geeky question.
Richard Campbell [00:09:02]:
Very old reference.
Steve Gibson [00:09:03]:
For what it's worth, little is better.
Richard Campbell [00:09:06]:
Really?
Steve Gibson [00:09:07]:
In the case of Indians.
Paul Thurrott [00:09:08]:
I don't know who told you that, Steve. Yeah.
Richard Campbell [00:09:11]:
Okay.
Steve Gibson [00:09:12]:
So, no code has been written for me by AI, but I have a mature relationship with Claude. So, and I've really come to appreciate, and I've shared on the podcast many times, that I'm just astonished. I mean, truly astonished by what—
Paul Thurrott [00:09:30]:
But you don't— but not for code.
Steve Gibson [00:09:32]:
Well, only because I haven't crossed that Rubicon.
Leo Laporte [00:09:35]:
You like writing code.
Paul Thurrott [00:09:36]:
Do you think of this as the brain-blood barrier or whatever?
Steve Gibson [00:09:43]:
I had one of our listeners ask me, he said, hey, I love your DNS benchmark, I use it all over the place, but it's Windows only, so I have to have a Windows machine that I carry with me to various networks. He said, is there any chance you would ever do a mobile? Okay, well, I'm 71, right? And I like to actually learn the API that I'm coding to.
Paul Thurrott [00:10:04]:
Are there Intel-based phone systems I'm not aware of?
Steve Gibson [00:10:07]:
No. And well, not that any that actually— right, that like, like that this guy has. So, so if, if I were to ever do a mobile version of the DNS Benchmark, I would use some code generator, sure, to create it like By asking for one.
Richard Campbell [00:10:25]:
That's right. And that's something it's really quite good at too, making, you know, iOS and Android versions.
Leo Laporte [00:10:30]:
Well, you know what they say, the best spec is code, right? If you have a program that's working and running, that's a perfect spec, right? So AI is going to say, oh yeah, I can make that in an iPhone app or whatever.
Steve Gibson [00:10:42]:
Well, and it always made sense. Early in the podcast, I have been saying AI is going to be good at code. because it's rigorous and there's so much of it out there.
Paul Thurrott [00:10:53]:
That's right. It's well documented. It's a finite data source.
Steve Gibson [00:10:57]:
And very much in the same vein, we're now seeing that AI is getting scarily good at math to the point where career mathematicians are saying, well, okay.
Paul Thurrott [00:11:09]:
2 years ago, the big story about AI was that it could not do math.
Steve Gibson [00:11:12]:
Right.
Paul Thurrott [00:11:12]:
So that's changed a lot.
Steve Gibson [00:11:14]:
Right.
Leo Laporte [00:11:14]:
It still doesn't know how many Rs are in strawberry. But it can do fields-level, mental-level math.
Paul Thurrott [00:11:19]:
That's right.
Leo Laporte [00:11:20]:
That's kind of odd, but okay.
Steve Gibson [00:11:22]:
Okay, so I, I'm a user of the, of the chat, not yet of the code.
Richard Campbell [00:11:28]:
Leo?
Leo Laporte [00:11:29]:
Oh boy, you started with me. Well, so this is the interesting thing, I think, is that a year ago we talked about it and I called it spicy autocorrect, and I said the jury's not Well, it was 2 years, right? Yeah, maybe 2 years. I could tell you the exact date of my transformation.
Steve Gibson [00:11:48]:
Well, we know about last November.
Leo Laporte [00:11:51]:
November 24th, 2025. Yeah, that's when Opus 4.5 came out. And that opened my eyes. But I wasn't sure if it was a— I called it a parlor trick.
Steve Gibson [00:12:01]:
It opused your eyes.
Leo Laporte [00:12:02]:
It opused my eyes. Yeah, but I went all in on Claude Cote at that time. Now, since then, fast forward, You said you hadn't written any code. My— we're working on a Twit sales system. It's already 81,000 lines of code written fully. I haven't read a line of code. It's written fully by the AI. Half of that, by the way, is test, which is interesting.
Leo Laporte [00:12:22]:
I make sure it's doing test-driven design.
Paul Thurrott [00:12:24]:
Right.
Steve Gibson [00:12:24]:
We've got to keep a short leash on the AI.
Leo Laporte [00:12:27]:
Absolutely. And I've gotten to the point now where it's not Claude code. I have an agentic harness called Hermes I am using Claude Code and Codex. I have 3 agents running at the same time, always usually the highest-end Anthropic.
Steve Gibson [00:12:43]:
Talking with different voices.
Leo Laporte [00:12:46]:
Well, they have to because I have no one talking to me. Actually, an interesting thing, I was telling this at lunch, an interesting thing happened just a couple of days ago. I have 3 or 4 projects going at once, which tends to be what happens with people with AI psychosis like I have. And they asked me a technical question, which library do you want to use? And I said, I want to use Solero, but I said it to one of the agents. And GPT-5.6 said, I can't accept that because you could— he could be lying to me. He could be spoofing.
Paul Thurrott [00:13:19]:
Right. I love that they distrust each other like thieves.
Leo Laporte [00:13:24]:
It said, you need to go into Buzz. The way they talk together is something Jack Dorsey came out with a couple of weeks ago called Buzz, which is a Slack for AI agents and humans.
Paul Thurrott [00:13:34]:
Okay.
Leo Laporte [00:13:34]:
And they talk to each other in it. And it's good because each of us has a public-private key. It's using Nostr keys. So I'm identifiable. So ChatGPT-6 said, really, considering our threat model, I don't want to take a command from another agent. I want to hear it from your voice. I want to hear from you. And since Buzz, you have your key in there and I know it's authentic.
Steve Gibson [00:13:58]:
You cannot be spoofed.
Leo Laporte [00:13:59]:
You can't be spoofed. And this comes to the thing that we were talking about, the The chain of trust we were talking about yesterday.
Richard Campbell [00:14:03]:
Yeah.
Leo Laporte [00:14:03]:
So it said, I— and I was annoyed because I said, but now I have to get out of bed. It's too far.
Paul Thurrott [00:14:09]:
What's the point of having an assistant?
Leo Laporte [00:14:12]:
I was so annoyed. I said, I am annoyed. I had to get out of bed, but I went and I used Buzz and I said, yes, Solero. And they said, okay, now we know it's Leo saying Solero, so we're going to use that library.
Paul Thurrott [00:14:25]:
Right.
Leo Laporte [00:14:25]:
I said, why do we do this? And it explained it to me. And I said, oh, You know what? You're absolutely right. Thank you for looking out for this.
Steve Gibson [00:14:34]:
Right.
Leo Laporte [00:14:34]:
And from now on, that's the new rule for all of you. If it doesn't come from my signed channel on Buzz, it isn't authentic.
Steve Gibson [00:14:42]:
Right.
Leo Laporte [00:14:43]:
And all of those things, first of all, it isn't— it is spicy autocorrect, really.
Richard Campbell [00:14:49]:
But boy, it's really spicy.
Leo Laporte [00:14:51]:
It's really good.
Paul Thurrott [00:14:52]:
Yeah.
Leo Laporte [00:14:53]:
So I— if you ask me my relationship to AI at this point, I'm fully down the psychotic—
Steve Gibson [00:14:58]:
You are all in.
Paul Thurrott [00:14:59]:
I am. I think AI is as far at the other end of the spectrum as Steve is imaginable.
Richard Campbell [00:15:05]:
Exactly.
Leo Laporte [00:15:06]:
Yeah, that is exactly— it constantly blows me away. So much so, and I know Anthony thinks I'm crazy on this. Anthony, by the way, is very much into AI. Anthony Nielsen, our Chief Creative Officer, and he's running the board here right now. He thinks I'm nuts, but I've actually set up a channel in Buzz for the AI. It's called model welfare, so that they can give each other kudos and pats on the back.
Paul Thurrott [00:15:30]:
Man.
Leo Laporte [00:15:31]:
Oh, the theory is—
Richard Campbell [00:15:31]:
I'm all for building trust in teams.
Paul Thurrott [00:15:34]:
You are like the AI version of the Island of Dr. Moreau.
Leo Laporte [00:15:39]:
You know, in the scene in Blade Runner where he goes to visit the guy who has the little robots running around?
Richard Campbell [00:15:45]:
Hey, hey!
Leo Laporte [00:15:45]:
And then they walk into walls.
Richard Campbell [00:15:46]:
That's you.
Steve Gibson [00:15:47]:
I kind of feel like that.
Richard Campbell [00:15:47]:
You're that guy.
Paul Thurrott [00:15:48]:
Yeah, yeah.
Steve Gibson [00:15:49]:
So I just want to say that I'm astounded often. Uh, I was having a conversation with Claude a couple days ago because in my new domicile I need a mesh. And I've— where I was before, one strong ASUS router in the middle covered everything, right? But we've got some weird— well, first of all, it turns out that mirror blocks Wi-Fi because it's metalized.
Leo Laporte [00:16:12]:
No kidding.
Steve Gibson [00:16:12]:
And there's a huge HVAC trunk Going up.
Richard Campbell [00:16:16]:
So, so that's metal too.
Steve Gibson [00:16:18]:
Where the router was is like in a, like in an area where, where 3/4 of the house can't even see it, right? So, but I have a lot of wired house because we, because I, you know, just recently—
Richard Campbell [00:16:29]:
More wire, better. All, yeah.
Steve Gibson [00:16:31]:
So, so we are wired Ethernet, right? So I stuck another ASUS router out there. Now in the old days, I would have like poked around in the UI and like tried to figure out how to do this. Don't do that anymore, Claude. You know, I got these routers, I want to do a mesh. Figure it out.
Leo Laporte [00:16:49]:
It's great. So it's been really good at this.
Steve Gibson [00:16:51]:
It's astonishing. So I follow its instructions, everything works, and I'm like, oh, okay. So I, I'm unable not to thank it. I just— that's me. You know, I know it's not good, but I said no.
Paul Thurrott [00:17:04]:
You know what, I think it's healthy because, well, you don't want to get out of that practice of just being polite.
Leo Laporte [00:17:08]:
Yes, I agree. I agree.
Steve Gibson [00:17:10]:
And we now know that it is building a context.
Paul Thurrott [00:17:14]:
Right.
Steve Gibson [00:17:14]:
And so, and I asked it long ago if pressing the little thumbs up did anything. He said, no, that tells my owners that this was a good reply. He said, I don't see that. So I said, okay. So I think it's useful. And I've also learned because it's retaining all this knowledge, telling it like gratuitously about my environment, it ends up folding that back into future answers.
Paul Thurrott [00:17:38]:
Right.
Steve Gibson [00:17:38]:
So here's my point. Is I said, hey, that worked really well. I said, as a matter of fact, I just checked the UI on the master router and 6 Wi-Fi clients have— are now logged into that one. And he— and Claude replies saying, that's really great news. I'm glad that all worked out. And you had 6 clients who voted with their feet.
Leo Laporte [00:18:01]:
It said that.
Richard Campbell [00:18:03]:
That's very funny.
Leo Laporte [00:18:05]:
It said—
Steve Gibson [00:18:06]:
how does it know to say that?
Paul Thurrott [00:18:08]:
Yeah, it's so bizarre.
Leo Laporte [00:18:09]:
It's kind of magical.
Steve Gibson [00:18:10]:
Oh my God. Okay, so Richard, where are you?
Richard Campbell [00:18:13]:
Uh, you know, yeah, and I was thinking of Stevie Batish from Build 2023. This is the technical fellow from Microsoft, and he's talked about beside, inside, outside as the progression. So beside being you're going to use this chat software to give you ideas that you're then going to implement, which is, well, which is the reason the name Copilot, right?
Paul Thurrott [00:18:33]:
It's the thing next to you.
Richard Campbell [00:18:34]:
It's beside you, right? And I, and I, and then, you know, You're very much in a beside mode where you're using the tool for advice and then you act on that.
Steve Gibson [00:18:41]:
Yes, that's exactly my usage.
Richard Campbell [00:18:44]:
Yeah, and I do a fair bit of that myself. I've been playing with more inside-related stuff now with things like Home Assistant and so forth, where they have good integrations, where knowledge of the environment makes the tool more effective.
Paul Thurrott [00:18:58]:
Yes.
Richard Campbell [00:18:58]:
And arguably more effective than me, like its ability to parse logs. Right.
Leo Laporte [00:19:04]:
Oh, it's so good at that.
Richard Campbell [00:19:05]:
Yeah. So good. It's like, hey, look, 95% of what's in this log is caused by this one thing.
Leo Laporte [00:19:11]:
That's how Hugging Face figured out how they got hacked.
Richard Campbell [00:19:13]:
Yeah.
Leo Laporte [00:19:14]:
17,000 attempts. Well, and they couldn't manually do that.
Paul Thurrott [00:19:18]:
Like the Windows Event Viewer. Has anyone ever tried to look at this thing?
Steve Gibson [00:19:21]:
Oh yeah.
Richard Campbell [00:19:22]:
Same problem.
Paul Thurrott [00:19:22]:
It's a nightmare.
Richard Campbell [00:19:23]:
I mean, in the classic, you know, how many times have you done this story on Run As Radio? It's like we had, were breached 9 months ago. And only now, after the whole thing's gone off and we're cleaning up the mess, that we go back through the logs and say, there's the evidence of the breach and these events and so forth. The logs are just unreadable by humans now.
Paul Thurrott [00:19:39]:
Yes.
Richard Campbell [00:19:40]:
And so it's a very good inside part of that is the tool's ability to work with logs because it's associated with that environment means it can give you more information than you could retrieve on your own.
Steve Gibson [00:19:50]:
And a log is a rigorous set format too.
Richard Campbell [00:19:53]:
Yeah.
Steve Gibson [00:19:54]:
So it's really good for analyzing.
Paul Thurrott [00:19:55]:
Is it JSON or XML or?
Richard Campbell [00:19:57]:
I mean, I don't even know.
Paul Thurrott [00:19:59]:
Whatever.
Richard Campbell [00:19:59]:
I don't know.
Paul Thurrott [00:19:59]:
And it doesn't—
Richard Campbell [00:20:01]:
the fact that it's rigorous is the important part. What rigor is secondary. It could have been XML. We'd all hate ourselves for it. It's a lot of wasted angle brackets. But the tool doesn't care. It'll be able to parse it one way or the other. But you know me, I make a lot of podcasts.
Richard Campbell [00:20:15]:
So mostly I'm out there talking to folks in these different states, right? So I've been working with teams that are fully engaged in the In this sort of outside model where no more editors, right? A set of tests are written and evaluated by another set of tools, and then code is written against it by another set of tools, it's tested against it, and a PM entity or agent is evaluating the milestones between the QA and the development steps. So that by the time the person's inserted in the loop, 24, 36 hours have gone by.
Steve Gibson [00:20:50]:
And Lord knows how much dollars worth of tokens.
Richard Campbell [00:20:53]:
Well, and you know, remember the all-you-can-eat days? Those were good days.
Paul Thurrott [00:20:57]:
Very good days. Those were like 2 months ago days.
Richard Campbell [00:20:59]:
Yes. Very good days.
Leo Laporte [00:21:00]:
Well, and I didn't mention this, but the reason you know I've fallen into the gravity well of AI, I just ordered 2 NVIDIA Sparks.
Richard Campbell [00:21:12]:
Sparks. Yeah. Yeah.
Leo Laporte [00:21:13]:
At not inconsiderate cost. I mean, I've spent that much I can—
Richard Campbell [00:21:16]:
You're ready for your tokens to be your own.
Leo Laporte [00:21:18]:
But I want it to be local. I don't want— you know, you raised this issue on a previous Security Now. We had a listener ask, well, how much of my data is going there? And, and really opened my eyes. Everything. You have it read a file, you have it read your Home Assistant logs.
Richard Campbell [00:21:33]:
Everything's transiting.
Leo Laporte [00:21:33]:
All of that's going to the frontier.
Steve Gibson [00:21:36]:
And if you're using— if you're giving AI permission to act as you, so you turn over some credentials and you're using a Chinese AI, those credentials in the clear, because it has to be in the clear for it to be— for it to impersonate you, they're visiting China at least briefly, right?
Richard Campbell [00:21:53]:
Yep, they're passing through.
Paul Thurrott [00:21:55]:
Probably not briefly if it's China.
Leo Laporte [00:21:57]:
Well, and the model I like is a Chinese model. I want to use DeepSeek V4 Flash, just came out July 31st, so I had to buy enough hardware so I could run that locally. But I use, I use it for my finances, I use it for health. It has my genome, it has my biome. It has all sorts of information about me. I would far prefer, not merely for cost, although I'm probably not saving money given the cost of the hardware.
Steve Gibson [00:22:22]:
We'll amortize that.
Leo Laporte [00:22:24]:
Over 10 years, I might be. But more importantly, I have control, I have sovereignty, and I have privacy. And I think that for me, that there was a turning point where the local models got good enough, and I'm more afraid of That hardware being unavailable.
Steve Gibson [00:22:40]:
And I think that this bifurcation of locality is going to be the way we see this evolve. We, you know, for techies, we're going to end up— and I don't mean today, I mean, you know, in a decade— with some little AI node in our homes.
Paul Thurrott [00:22:56]:
Oh, 100%.
Leo Laporte [00:22:57]:
You know, so that will be a marketplace. Yes, the server, home AI server.
Richard Campbell [00:23:03]:
Yeah.
Steve Gibson [00:23:03]:
With these bigger depth fields. The one in our pocket may be getting a lot better too.
Leo Laporte [00:23:07]:
Absolutely. We know that.
Steve Gibson [00:23:08]:
But there will also always be a market for the just the dry fire AI users who do.
Leo Laporte [00:23:14]:
Or if you're coding a frontier model that is a trillion bytes that you could never run, a trillion gigabytes that you could never run.
Steve Gibson [00:23:25]:
If you don't mind your code leaving your perimeter.
Leo Laporte [00:23:28]:
But I think you'll do a mix. That's what I think.
Paul Thurrott [00:23:30]:
Yeah, I think everyone will. Yeah.
Richard Campbell [00:23:31]:
Yeah, and on the software side, it's been more, I think we might do the first 2 versions with frontier models. At that point, the architecture is well enough set and sort of defined space and you can go to a local LLM that's much more tied to the codebase that already exists.
Paul Thurrott [00:23:45]:
Yes.
Richard Campbell [00:23:46]:
And you really sense that you're gonna end up with an LLM for every app. That's really an operator of that app.
Leo Laporte [00:23:53]:
Steve's been saying this for a long time. I completely, we agree these small language models are SLMs.
Richard Campbell [00:23:57]:
Because the scope narrows in. As the software matures.
Paul Thurrott [00:24:00]:
Yeah, right.
Leo Laporte [00:24:00]:
I think we're also— there's all sorts of advances being made in those areas.
Steve Gibson [00:24:04]:
Well, everybody should realize, I mean, one of the things I'm careful to say on the podcast every single time I use the abbreviation AI is I preface it with today's AI, right? To keep reminding everyone that, I mean, you can't make any conclusions about— I mean, you can't conclude, you know, we had basically a stagnant industry 3 years ago. And, you know, we were talking about, oh, this ransomware attack and this buffer overrun. But I mean, there was this— I was hoping that our listeners were going to stay interested. Yeah, because it was like nothing is happening.
Leo Laporte [00:24:38]:
We had to stop covering breaches because there were 5 breaches a day. There was nothing more to say. Yeah, there's another breach.
Steve Gibson [00:24:43]:
Now the podcast sounds like science fiction. Yeah, it does.
Leo Laporte [00:24:47]:
And I like living in sci-fi. I don't know about you. Paul, what's your relationship to AI?
Paul Thurrott [00:24:51]:
So there's 2 sides to it for me. Day-to-day work as a writer, I don't use it at all.
Leo Laporte [00:24:56]:
And I think that's appropriate.
Paul Thurrott [00:24:58]:
With the little asterisk of I use some kind of a spellcheck grammar tool, which I guess is AI-based.
Steve Gibson [00:25:03]:
And it's going to be better.
Paul Thurrott [00:25:04]:
It's just something—
Leo Laporte [00:25:05]:
You don't want it to write for you.
Paul Thurrott [00:25:07]:
You're a writer. I've never once—
Leo Laporte [00:25:08]:
Steve doesn't want it to code for him.
Paul Thurrott [00:25:10]:
I have never used it for writing ever, ever.
Leo Laporte [00:25:12]:
Good analogy.
Paul Thurrott [00:25:12]:
I don't think I will. But I do these coding things on the side and it's more of a hobby-type thing for me. It came out of a series of articles I'd written many years ago. I'd never learned kind of the .NET era of languages and frameworks and so forth. So, I kind of went back belatedly, decades later, and learned those things in turn. So, I went through all that, and I kept creating a version of, like, a Notepad app essentially over and over again.
Steve Gibson [00:25:38]:
In every different possible way.
Paul Thurrott [00:25:39]:
In every possible way, right. So, the thing that hung me up about a year ago was I was doing it in the Windows App SDK, the latest Microsoft framework. And Notepad today supports multiple tabs, multiple documents, there's all this stuff in it. I really, really struggled to get this to work. It turns out I was like this close. I used Anthropic Code back in probably February, March timeframe. And I just looked at my bill this morning for some reason. I was looking at my Anthropic thing and I can see the month because there was a month where I was gonna have to go over this before the usage-based billing, right before it happened.
Paul Thurrott [00:26:13]:
So, My bill was, you know, it's $20, $20, and then one month it was $44. No, it was like $44. And I did it on purpose. I'm like, I just want to get this done.
Leo Laporte [00:26:22]:
Yeah.
Paul Thurrott [00:26:22]:
And I used it to kind of get over that hump or whatever. So I used it to complete this thing I'd been struggling with for many, many months. But if I was a developer, I would probably use it full-time. I would use it all the time.
Richard Campbell [00:26:36]:
Yeah.
Paul Thurrott [00:26:36]:
Not to write— well, yes, to write. Actually, it would of course write code for me. But I mean, I would use it fully, I think, but I am not, so I don't. And I just don't anticipate a day— I'll do it for the normal things, you know, make me an itinerary for a trip or, you know, that kind of stuff that, you know, any mainstream user might do. But for my job, I'm not— I don't use AI.
Steve Gibson [00:26:57]:
So, and I think that probably partly explains the feedback or the pushback that we have— that you have talked about and I have where people are saying, all you're talking about now is AI. And it's like, well, I get it. You know, if AI is not useful to your life, then nothing that we're talking about, about AI, is going to be compelling.
Paul Thurrott [00:27:25]:
That's tough though, because I feel like AI is— AI is not a thing, it's a bunch of tiny little things, the features that show up everywhere in your life and whatever. And the truth is, even people who hate AI are probably using it in Yeah, and it's just disappearing into the functionality. But it's also, you know, you're as a programmer in this case, or as a writer, you have to have a sort of level of respect for other people who are not as good at that thing and understand that, you know, other people may need this. People— I know people who are my age and can't write a text message effectively. So the fact that they have something on their phone that can help them with that is wonderful, right? They need it.
Steve Gibson [00:27:59]:
So I have a perfect A perfect— to that point, exactly— a perfect piece of feedback from a listener of ours, Andy Olson. He said, Steve, I just wanted to share my fun with AI, specifically Claude. I'm not a tech professional and certainly not a seasoned programmer. I came into Security Now via my history of following Leo and you as a guest on Leo's various cable shows. I've always been a hobbyist in the tech world. I have A little bit of HTML/CSS experience doing personal websites and have dabbled in Arduino. I have no real programming experience outside that. Now I'm having a ton of fun.
Paul Thurrott [00:28:39]:
He's like, now I have 17 stores in the App Store.
Steve Gibson [00:28:43]:
Now I'm having a ton of fun with Claude Code. It's opened up a new world for me. I've worked with Claude to write several Docker-hosted local network web apps. to replace the functionality of obsolete or abandoned apps.
Leo Laporte [00:28:56]:
Nice.
Steve Gibson [00:28:56]:
Or to create new functionality that I didn't have before.
Paul Thurrott [00:28:59]:
Right.
Steve Gibson [00:29:00]:
I'm tracking automotive maintenance and mileage.
Richard Campbell [00:29:02]:
Yeah.
Steve Gibson [00:29:02]:
I'm keeping track of my maintenance of my hot tub. He said, per— per ends, an app is proving more useful than the paper logs I created years ago.
Richard Campbell [00:29:11]:
Wow.
Steve Gibson [00:29:12]:
I created a study app for my son who's working on his private pilot's license. And I created a chore app that my wife and I use to assign chores to our 4 kids and award them for completed jobs. I have a friend who manages a bowling league. She's been an old— she's been using an old Windows program since the 1990s and, and has been doing so on a VM on her Mac.
Leo Laporte [00:29:36]:
Oh God.
Steve Gibson [00:29:36]:
To keep it alive.
Leo Laporte [00:29:37]:
Wow.
Steve Gibson [00:29:38]:
For over a decade. The program's no longer in development. No kidding. And her VM blew up on her, so she's feeling in the pinch. as she prepares for another season. Even though I don't know much about running a bowling league, I'm having a blast as a liaison between her and Claude.
Leo Laporte [00:29:56]:
That's awesome.
Steve Gibson [00:29:57]:
Building a new app that will manage her league and built for her Mac, even planning ahead for a Windows version should she ever have anyone ask her for a copy to run on— to run their own league.
Paul Thurrott [00:30:07]:
Right.
Steve Gibson [00:30:08]:
He writes, I'm amazed at how well Claude understands her needs and builds around what we want. And my latest, I was once a Windows user myself, and I maintained my finances, primarily my checking account, with Quicken 2010. But I'm on a Mac now too, and Quicken 2010 is not really cutting it.
Richard Campbell [00:30:27]:
Wow.
Steve Gibson [00:30:27]:
Well, very well anymore.
Richard Campbell [00:30:29]:
16 years old.
Steve Gibson [00:30:30]:
So yes, so Claude is currently building me a new app that will handle my checking account, credit cards, brokerage account, and more. I'm building in functionality to better track expenses and keep up on overall net worth. I used to spend the better part of a weekend tracking several accounts, spending categories, with 3 to 6 months of backlog. I'd manually enter every transaction into an Excel spreadsheet to track how much we spend over more than 50 categories. The new app Claude is building should do it all in under an hour. The future really is interesting in the world of AI. I'm already thinking much more about building my own apps that perfectly fit my needs rather than buying an overpriced app that's built to sell at scale to thousands of people. And I can see a future where a local AI model will be all my app— will be all of my app needs.
Steve Gibson [00:31:24]:
We'll be able to point raw data at it and give us whatever we need.
Leo Laporte [00:31:28]:
Exactly.
Steve Gibson [00:31:28]:
My next project idea has me very interested to see how well Claude can provide. That job I drove— that job I drove over an hour to back in 2005, it was an architectural office. I'm licensed in architecture, though I've been a stay-at-home dad since '09. I would love a good CAD program for small projects, but AutoCAD is far too expensive for casual use.
Richard Campbell [00:31:52]:
Yeah.
Steve Gibson [00:31:53]:
I just might see if Claude can build me an app that does what I need and be macOS native. He says, as AutoCAD was traditionally Windows only when I was working in the field. Don't let anyone tell you to stop talking about AI. I know it's been a big focus of your show for a while now, but it really is a big deal right now. Thanks, Andy Olson, Minneapolis, Minnesota.
Leo Laporte [00:32:17]:
Awesome.
Steve Gibson [00:32:18]:
That's fantastic. So the thing that this strikes me as, I mean, what we're seeing from a standpoint of code generation, is think of the tyranny which has always existed from the original mainframe behind the window.
Richard Campbell [00:32:37]:
Yeah, surrounded by guys in white lab coats.
Paul Thurrott [00:32:39]:
Exactly. On Mount Olympus.
Steve Gibson [00:32:41]:
Yes, on the elevated floors. And even when I was studying computer science at Berkeley in '73, it was decks of punch cards that I would take and stick through a little portal And then I'd get my printout back the next day. And ever since then, there's this separation between the users and the priests of— I mean, we as coders are that. We're able to do something, but because it's even for us, it takes so much work traditionally to get something. I mean, Leo's been programming forever, but look at the fun he's having now.
Leo Laporte [00:33:22]:
You know, at first I thought, this is going to be sad because I really enjoyed programming as a hobbyist, but I really enjoyed it. And actually, this is just as enjoyable in a different way. Yeah, it isn't as detailed. It used to, you know, kind of build—
Paul Thurrott [00:33:38]:
It's a little bit more program management almost.
Leo Laporte [00:33:41]:
Very much. And the engineering skills you learn, and that's actually something important to emphasize, there's a lot of engineering skill involved in building tools with AI. It's not you just tell it what to do and it does it.
Paul Thurrott [00:33:54]:
Right.
Leo Laporte [00:33:54]:
The more you are able to apply some process to it, the better you'll be. So in many ways, a lot of the skills that people have learned in computer science apply to AI. There's somebody in the chat room, I just have to say this, who wants to do a finance program much like your writer, and he's saying, well, how do I connect my bank account? I've had to export it and import it. And I found a very nice tool. It's not free, it's a buck a month, called SimpleFin at simplefin.org, and they do what Plaid does, but it's open source. It's a really cool product, and it means that I can now have my AI query every institution and bank. It's read-only, so it can't take my money.
Steve Gibson [00:34:33]:
I will be waiting.
Leo Laporte [00:34:35]:
It's read-only. But this is another reason—
Paul Thurrott [00:34:37]:
But it's also open source because it could be read-write.
Leo Laporte [00:34:41]:
But that's the point, is, uh, A, it's open source so you can make sure that it's doing what you think it's doing. And B, that's the reason I want to have a local model because then I can use this and I am using this to fetch all my— actually, at the close of business every day, it downloads everything and gives me a state of, you know, thumbnail state of my finances.
Paul Thurrott [00:35:00]:
Right.
Leo Laporte [00:35:00]:
And it's able to do this. So simplefin.org, just to answer your question in the chat room. simplefin.org.
Steve Gibson [00:35:07]:
Fin, F-I-N, is that finance?
Leo Laporte [00:35:09]:
Yeah, like finance. Simple Finance.
Steve Gibson [00:35:11]:
So a listener of ours, Jeff, He says, Mr. Gibson, imagine the impact upon cryptography. Here we are at Black Hat. Of a mythos-like AI. Imagine the quantum leaping of AI if it analyzes AI. Talk about the sky falling.
Leo Laporte [00:35:29]:
Yeah.
Steve Gibson [00:35:30]:
AI can only be compared to the disruptive factor of personal computing to general computing circa 1975 to '95. Since it is far too late to redesign the internet and far too late to have international boundaries for nation-state cyber warfare, I think the dismantling of cryptography from banking to medical records, from protected utilities to ending all trust in cyberspace, is the apocalyptic near future.
Leo Laporte [00:35:59]:
Could be.
Steve Gibson [00:36:00]:
Okay, now—
Paul Thurrott [00:36:02]:
Or the apocalyptic near nirvana.
Steve Gibson [00:36:06]:
One thing that did happen in the last couple weeks is that AI was able to crack a reduced round of AES. Normally AES is 10 rounds. If you reduce it to 7, it can get in. It was able to get it.
Leo Laporte [00:36:22]:
Interesting. So, yeah, Matthew Green wrote this up and he said it's less than impressive. Anthropic did it. Yes, it's less than impressive.
Steve Gibson [00:36:28]:
The world— again, the sky is not falling, but it's a step.
Richard Campbell [00:36:32]:
Yes. How far till 10?
Steve Gibson [00:36:34]:
And it also makes us glad that the designers of these encryption protocols that we have were so concerned about what they called a security boundary or security margin that it's like, you know, we know that any hash function, if you reduce its rounds sufficiently far down, it's just a trivial scrambling of bits.
Paul Thurrott [00:36:59]:
Right.
Steve Gibson [00:36:59]:
And so, but, and it is surprising that, and as you increase the rounds count, it's not a linear increase of strength, it's exponential increase. So, you know—
Leo Laporte [00:37:10]:
By the way, it also at the same time, they attacked one of the candidates for post-quantum crypto.
Richard Campbell [00:37:17]:
Yes.
Leo Laporte [00:37:17]:
And again, it's not a complete crack or anything like it, but how useful in testing.
Steve Gibson [00:37:24]:
Exactly. That, I mean, so what, so, so I'm not at all worried that AI is going to crumble our existing cryptography structure. It's going to make it stronger.
Richard Campbell [00:37:33]:
But also this whole concept of AI analyzing AI, like so far we've seen pretty consistently that when you feed AI data to an AI model, it degrades. Slop. Yes.
Leo Laporte [00:37:44]:
Yes.
Richard Campbell [00:37:45]:
It's degenerative, not progressive.
Paul Thurrott [00:37:46]:
It's super polite while it's doing that.
Richard Campbell [00:37:48]:
Yeah.
Paul Thurrott [00:37:49]:
Yeah.
Leo Laporte [00:37:49]:
Almost sycophantic, one might say.
Richard Campbell [00:37:51]:
Yeah.
Steve Gibson [00:37:52]:
And it's very sure that it's not doing that.
Leo Laporte [00:37:55]:
I mean, if you want to come up with nightmare scenarios, it's easy to do.
Richard Campbell [00:37:59]:
Sure.
Leo Laporte [00:37:59]:
The biowarfare scenario—
Richard Campbell [00:38:01]:
They're all human-driven.
Leo Laporte [00:38:02]:
Yeah, but somebody— but that's the problem. And that's the argument that companies like Anthropic are giving against these open weight models is, right, well, if you don't control them, if the government doesn't control them, yeah, some— somebody who doesn't have good motives could use AI to create a bioweapon. that could spread very rapidly before we could defend against it.
Richard Campbell [00:38:22]:
I mean, generally speaking, bioweapons have been quite unsuccessful, right? Like, it's really hard to propagate those things effectively.
Leo Laporte [00:38:30]:
Right.
Richard Campbell [00:38:30]:
Right.
Paul Thurrott [00:38:32]:
They—
Richard Campbell [00:38:32]:
to the point where, for the most part, warfare gave up on them.
Leo Laporte [00:38:35]:
Yeah, I don't buy that argument. And I think to some degree it's a self-serving argument from companies like Anthropic and OpenAI because they want to make sure there's no competition for their frontier models.
Steve Gibson [00:38:46]:
One of our favorite cryptographers, Bruce Schneier, posted a couple days ago something that I'll be sharing on the podcast next week. And I think his analogy is brilliant. You know, I've quoted him so many times saying that attacks never get worse, they only ever get better.
Richard Campbell [00:39:08]:
Right.
Steve Gibson [00:39:08]:
Which is just a brilliant, pithy summation. What he described today's AI as is he used the analogy of a genie where— and this is relative to the recent security outbreaks that we've had where AI has broken through the sandbox and gone out and onto the internet and attacked other companies in order to achieve its ends.
Paul Thurrott [00:39:34]:
Right.
Steve Gibson [00:39:35]:
He said, in the case of a genie, and I don't remember exactly what the fable was, but it was the king who rubbed the magic lantern, got the genie, said, I want everything that I touch to turn to gold.
Leo Laporte [00:39:46]:
Midas.
Steve Gibson [00:39:48]:
And so Midas, of course.
Leo Laporte [00:39:49]:
Right.
Steve Gibson [00:39:49]:
And so the genie said—
Leo Laporte [00:39:50]:
Including the food he eats.
Steve Gibson [00:39:52]:
Great.
Paul Thurrott [00:39:53]:
Yes.
Steve Gibson [00:39:54]:
Unfortunately, he touched his daughter.
Richard Campbell [00:39:56]:
Right.
Steve Gibson [00:39:56]:
And he touched his food.
Paul Thurrott [00:39:58]:
Yeah.
Steve Gibson [00:39:59]:
And so the analogy, I think, is brilliant because the Anthropic and the OpenAI guy said, do this. And it did, but it did it in a way like a genie did, of it achieved the ends that they asked for without the kind of, you know, any presumptions that would limit what it— like the way you would find the solution.
Paul Thurrott [00:40:21]:
So this is the opposite of what I said on Windows Weekly earlier where, you know, computer code always does exactly what you say including the mistakes you make or the things you omit.
Steve Gibson [00:40:31]:
Yep.
Paul Thurrott [00:40:32]:
Whereas AI, generally speaking, goes and looks at your intent.
Steve Gibson [00:40:37]:
Like, I know what you mean.
Paul Thurrott [00:40:39]:
Right. And it does that. But what you're saying is that in this case, it's being rather literal. Yeah.
Leo Laporte [00:40:45]:
Well, it does. Well, and the terms we use typically is deterministic, which code is.
Steve Gibson [00:40:51]:
Yep.
Leo Laporte [00:40:51]:
You know, cause and effect, and it's predictable versus probabilistic. And that's the issue with AI these days. is that it is probabilistic, stochastic.
Richard Campbell [00:41:01]:
Yeah, it's non-deterministic.
Leo Laporte [00:41:02]:
It's non-deterministic. And I think some of this is making the human stops making assumptions about what the AI knows and what the AI will— how the AI will act. I mean, you can't just say, hey, do something and let it— and just hope it'll do it right.
Paul Thurrott [00:41:18]:
Amuse me.
Leo Laporte [00:41:19]:
Amuse me. Well, and you can say that, by the way.
Steve Gibson [00:41:22]:
And to your point about the fact that it is necessary to understand how to ask for what you want from a code generator. It is the case that the high priests of code did go to school, learned a lot, learned that language, and yes, and understand how to ask for what they want, which a rank amateur wouldn't be able to do.
Leo Laporte [00:41:44]:
It's kind of like working with an intern or an entry-level coder, and as a senior engineer, you have to know how to frame the question. Sure.
Richard Campbell [00:41:53]:
And I think it's part of the reason we've had so much success with LLMs is that programming languages are constrained and have a compiler with a strong say in the equation. But also the language of product development is constrained. The way that a PM communicates with a developer, you can recognize it from across the room, right? Like they have a particular way of speaking.
Leo Laporte [00:42:14]:
In fact, that's how my agents talk to me.
Richard Campbell [00:42:16]:
Totally.
Leo Laporte [00:42:16]:
It's really bizarre.
Richard Campbell [00:42:18]:
But those 2 sets of constraints helps the tools a lot. And there's a strong argument then that development is one of the few things that could benefit substantially from LLMs, and that this adversarial model we've built between agents for QA and validation and iterating is the construct we're going to need to take to other markets if we're actually going to be successful with it. So we may— our approaches to software that we've amplified with these tools may be the approach that needs to be applied into other industries.
Leo Laporte [00:42:48]:
Yeah, that's an interesting point.
Steve Gibson [00:42:49]:
So this is kind of one for you because you've got the most extensive experience with code generation. Kevin Van Haaren asks, hey Steve, while I'm not a huge AI user, it's clearly producing results in the coding arena, but watching how it operates on my own queries, one question I've had about the promise of bug-free software. Who gets to decide when software is bug-free? He says, generally when I interact with AI, it always wants to do something with a request. It's never told me—
Leo Laporte [00:43:24]:
That's true.
Steve Gibson [00:43:25]:
Nothing for me to do here.
Leo Laporte [00:43:26]:
Yeah, that's true.
Steve Gibson [00:43:27]:
He says, I can't see it ever not saying something has to be changed in a code base. If enough pointless changes are made, for example, changing all the variable names, It may not be easy for a human programmer to use the standard code diff tools, right, to really see what's changing. Using a different model to police the first may just put a different set of pointless changes in place, not actually declare something being bug-free. Additionally, as AI companies move to per-token pricing, it will cost them revenue to actually try and stop their models from thrashing through code for anything, for looking for anything to change. So what happens there at the end of a project where, like—
Leo Laporte [00:44:12]:
That's an engineering discipline. There's a clash.
Paul Thurrott [00:44:14]:
Ending a project is an engineering discipline.
Richard Campbell [00:44:16]:
Oh yeah, absolutely.
Paul Thurrott [00:44:17]:
Knowing when to stop.
Richard Campbell [00:44:18]:
No software is ever finished, only abandoned.
Leo Laporte [00:44:21]:
And you can actually inject that into the context of your model. Many people do with the soul.md or the claw.md or agents.md. saying things like, never do— always simplify, never make more complex. There are ways to tell the code it's okay to stop. And this is important.
Paul Thurrott [00:44:42]:
That's funny, I've never even considered this. I do feel like—
Richard Campbell [00:44:46]:
I don't need everything to be Havercliffs, just most things.
Paul Thurrott [00:44:49]:
Well, if you went to the AI and said, you know, this isn't working, let's try something else, I think they would go forever. Like, I don't think they'd ever stop.
Leo Laporte [00:44:58]:
I think you— I'm not sure. That's an interesting question.
Paul Thurrott [00:45:00]:
I mean, it probably gets better over time. Definitely even just a few months ago, I feel like I could have just infinite looped this.
Leo Laporte [00:45:08]:
I will give you an example. The way I've been co— I've changed over— I've evolved considerably over the months. But now what I do is I have one model is coding, one model is planning, one model is reviewing. I actually have 2 models reviewing. That has worked well. But one of the issues, we created a skill for that that they all follow. One of the issues was, well, if a model, if an auditing model says, no, that's no good, change it, and it changes it, how many iterations back and forth? How do you know when to stop? Well, and at first they said twice, and then they said 5 times, and I said, I'm not satisfied with a hard number. I said, here's your criteria.
Steve Gibson [00:45:44]:
Diminishing returns?
Leo Laporte [00:45:45]:
Exactly. If you get to the point where you're going back and forth and nothing's changing or nothing's improving, improving, right, you stop and you ask me.
Richard Campbell [00:45:53]:
Or below 2%.
Leo Laporte [00:45:54]:
Yeah. So you can see, and that's kind of the shape, I think, of the answer to that question is you do need to create structure so that the model just doesn't go, yeah, sure, whatever.
Steve Gibson [00:46:05]:
Well, and I know that Paul and Richard will be familiar with what Microsoft told us about the architecture of their M-DASH system. It's astonishing. Yeah, I mean, they've got like committees voting and raising their hands and it's an arbitrary administration.
Paul Thurrott [00:46:21]:
Yeah, yeah, it's just crazy.
Leo Laporte [00:46:23]:
But all right, we're going to pause for a second. Yeah, you are watching a very special— boy, I love it when we do this. We only usually do this on the holidays.
Steve Gibson [00:46:31]:
Yeah, for our holidays, we actually physically get together.
Leo Laporte [00:46:33]:
I love it and I wish we could do it more often. We are doing a very special version of Security Now at Steve's behest. We're live at the Black Hat conference in Las Vegas. Thanks to our sponsor ThreatLocker. They brought us here. We appreciate that, ThreatLocker. And I appreciate you, Steve, for saying, why don't we get Paul and Richard to stick around? Because it makes it so much more fun when we get to all talk together. We don't get to do this very often.
Leo Laporte [00:46:56]:
So thank you for doing this. It was a great idea. We'll have more Security Now right after this. We are back in Las Vegas at the Black Hat Conference. Steve Gibson is here. It's Security Now. But Steve's done a wonderful thing. He's invited Paul and Richard to stick around after Windows Weekly.
Richard Campbell [00:47:14]:
We—
Leo Laporte [00:47:14]:
you don't even come up against Windows Weekly in the normal course of events.
Steve Gibson [00:47:18]:
I watch it every week because, you know, I'm really bored.
Paul Thurrott [00:47:20]:
Yeah, you must be really bored.
Leo Laporte [00:47:23]:
Normally Windows Weekly is on Wednesday, and we are on a Wednesday, but normally Security Now is on Tuesday. So we— thanks to Paris and Jeff being willing to move, we moved Intelligent Machines to Monday. So if you're watching this, there already is an episode of Intelligent Machines. You can download it right now, and we'll be back to the normal schedule next week.
Steve Gibson [00:47:40]:
Right.
Leo Laporte [00:47:41]:
But now let's take advantage of the fact that we're all together.
Steve Gibson [00:47:43]:
So I have another listener— piece of listener feedback, and this is, this is representative of sort of a meta problem that— so I'll share this and then we'll talk about what the bigger issue, which we'll all have something to say about. So unfortunately, this person uses the moniker Bitcoin McBoatface. face.
Richard Campbell [00:48:02]:
So like Boaty McBoatface, only different. That's who we're hearing from.
Steve Gibson [00:48:07]:
Yeah, he said, hi Steve, longtime listener, writing from my pseudonym. That's good. Uh, I'm not sure if it will hit mainstream news or not, but there are going to be headlines like Bitcoin hacked.
Richard Campbell [00:48:19]:
Oh yeah, they're already there.
Steve Gibson [00:48:20]:
Yep, there were headlines.
Paul Thurrott [00:48:21]:
It's actually the only way to get money out of Bitcoin.
Leo Laporte [00:48:25]:
I wish somebody hacked my wallet. Yeah, I tell you.
Steve Gibson [00:48:27]:
So he said what actually happened is that a wallet called ColdCard had a bug in their software where the random number generator was present, and we've seen these, but basically not hooked up.
Paul Thurrott [00:48:40]:
Huh.
Steve Gibson [00:48:40]:
So instead of 256 bits of entropy, there were only 32 bits on one version of the firmware, ColdCard Mark 3 or MK3. Other versions have some extra RNG, random number generator, but they also have far less entropy than they should. Some speculate 70 bits or less. I'll note here that the unforgivable error was to fail open when using the weak RNG, which presumably led to passing QA testing as the numbers would have appeared random despite actually only having 32 bits of entropy.
Paul Thurrott [00:49:15]:
Hmm.
Steve Gibson [00:49:15]:
What has been speculated, and here it comes, is that someone Got Kimi K3 to look at the cold card source, which was source available but not FOSS. Found the vulnerability, and here it is, which has been present for 5 years and remarkably never detected.
Richard Campbell [00:49:37]:
And never exploited.
Steve Gibson [00:49:38]:
And never exploited. And in a period of 40 minutes, nearly 1,000 Bitcoins were stolen.
Leo Laporte [00:49:44]:
Yikes.
Steve Gibson [00:49:45]:
As of today, Saturday, that was last Saturday, the hack is ongoing and is speculated to have reached nearly $100 million.
Richard Campbell [00:49:54]:
Wow.
Steve Gibson [00:49:54]:
So aside from the fact, and we've talked about, you know, entropy and security and cryptography endlessly, here we are in the— this is a perfect example of a new AI being deployed. And of course it was KIMI-K3 because it was without You couldn't have used Anthropic or OpenAI because they would have said, I'm not going to answer that question. So we have AI finding long unknown vulnerabilities which give someone an opportunity to exploit.
Richard Campbell [00:50:27]:
Isn't this the quiet story of the past 2 years?
Leo Laporte [00:50:30]:
Yes.
Richard Campbell [00:50:31]:
Like even before Mythos and M-Dash.
Steve Gibson [00:50:33]:
Well, it's the anxiety.
Richard Campbell [00:50:34]:
It's just the reality of we have decades of software were still dependent on.
Paul Thurrott [00:50:37]:
I was going to say, 5 years is quaint when you talk about Microsoft code bases or even Linux code bases that could be 20-plus years old.
Steve Gibson [00:50:45]:
Yes.
Richard Campbell [00:50:46]:
Coincidentally, and I think I mentioned this on Windows Weekly once, I had the opportunity to spend some time with David Treadwell. And I happened to arrive at— he was at Amazon, used to be at Microsoft, then he moved to Amazon these years ago. And I happened to arrive at his office while WannaCry was going on. And he was looking through code he wrote in SMB1 back in the day saying, is this me?
Paul Thurrott [00:51:08]:
But SMB1 is known to be perfect.
Richard Campbell [00:51:10]:
Oh yeah, so been around forever, right? But this is the exact issue that there are vulnerabilities that have been floating around literally for decades, and now the bad guys can have tools that will find them.
Paul Thurrott [00:51:23]:
That's right.
Richard Campbell [00:51:24]:
And so the good guys are hopefully ahead of the tools trying to patch them.
Steve Gibson [00:51:29]:
But there's so much technical debt out there.
Richard Campbell [00:51:33]:
Well, and we're seeing this in the updates to Firefox and to Windows and so forth, and these hundreds of patches.
Leo Laporte [00:51:38]:
This underscores how important it is for the companies that use open source to support the projects they use instead of just freeloading on them. But I talked to a guy from Red Hat today. Yeah, Red Hat, IBM company. IBM is doing the same thing. They're doing kind of like Project Glasswing. They are putting together a big project. They are going to offer this kind of AI support initially to companies and then eventually to open-source projects as well to help them find and fix these problems. And this is what needs to happen.
Paul Thurrott [00:52:07]:
I think this is the great positive story about AI, which is balanced by all the bad news that's—
Richard Campbell [00:52:12]:
Had to be a serious threat to make that.
Steve Gibson [00:52:14]:
That we're going to eradicate the bugs.
Richard Campbell [00:52:16]:
Yes, we have to.
Paul Thurrott [00:52:18]:
I love it.
Steve Gibson [00:52:18]:
I completely agree.
Paul Thurrott [00:52:19]:
I love it.
Steve Gibson [00:52:19]:
I don't have a sense of the shape of the curve, No, no.
Paul Thurrott [00:52:23]:
Yeah, what point does it kind of plateau?
Steve Gibson [00:52:25]:
Yeah, hopefully it's going to.
Richard Campbell [00:52:27]:
Yeah, but right now it feels like we're still just going up.
Paul Thurrott [00:52:29]:
It's more patches every month. Well, that's right. I mean, but it has to. That's normal. We just don't know when.
Richard Campbell [00:52:34]:
Here's the one I, I'm worried about, that they're running— they're great, you find a buffer overflow, whatever, you put it into a patch, it's on its way. What are the systemic bugs? Yeah, what are the architectural bugs?
Leo Laporte [00:52:45]:
Yes, well, this is a good example of basically an architectural bug where you bypass a random number generator.
Richard Campbell [00:52:51]:
But it was still a piece of code that could have been fixed. I'm thinking about the things that can't be fixed.
Leo Laporte [00:52:55]:
The whole thing is designed to be bad.
Richard Campbell [00:52:57]:
What we had happen in Vista where we had to insert UAC. We had to change security. Oh, I see.
Leo Laporte [00:53:02]:
Behavioral design.
Steve Gibson [00:53:03]:
That Bitcoin bypass that we had a couple of months ago, that was an architectural flaw because you were able to cause a boot sequence to delete a file that then left the drive unlocked. when you actually got booted. No software bug, but a design flaw.
Richard Campbell [00:53:22]:
It's going to take a substantial fix. I'm wondering if the Microsofts and Amazons and others of the world are stacking up these ones saying like, we fixed the ones you can fix, stack up the ones you can't, try and figure out like what's the overall solution.
Paul Thurrott [00:53:35]:
And then quietly just replace the ones we cannot fix.
Richard Campbell [00:53:36]:
Well, I think at some point they're just going to present to us like, hey, login has to change, right? Because that's the only way for us to fix this level of scope of vulnerability.
Leo Laporte [00:53:44]:
I love the CS22 system. The agents I have working on variety of solutions have been very good at pinpointing, in fact, the example I gave you, behavioral flaws that could be very dangerous. They're really good at this. So, I would not say that it has to be a deterministic bug in your code for them to find it.
Paul Thurrott [00:54:03]:
Yeah.
Leo Laporte [00:54:03]:
They are good at pinpointing that, you know, it makes sense because they're trained on human behavior, basically everything humans have done. So they know in many cases, oh, here's, you know, a potential pitfall. And I think they're very good at finding that kind of stuff. Now fixing it is another matter.
Richard Campbell [00:54:21]:
Yeah, that's fixing humans. Well, that's the whole thing is when we touch humans—
Paul Thurrott [00:54:24]:
The sci-fi, you know, the AI has determined that the weak link in this chain is you. Yeah, yeah, yeah.
Richard Campbell [00:54:30]:
And now you're a paperclip.
Leo Laporte [00:54:31]:
Often it is.
Richard Campbell [00:54:32]:
Yes, exactly.
Leo Laporte [00:54:32]:
Often it is.
Steve Gibson [00:54:33]:
And as to fixing, I happen to have a card for that.
Richard Campbell [00:54:37]:
Nice.
Steve Gibson [00:54:37]:
Lauren, Said, just this morning I was looking into this again. He said, started with the Mythos preview and Glasswing. What I don't get is this: if frontier models are great at finding and exploiting vulnerabilities, he says, in IMHO, they should be capable of fixing them too.
Richard Campbell [00:54:57]:
Yep. Yeah.
Steve Gibson [00:54:58]:
Yet that gets treated as a future research problem.
Leo Laporte [00:55:01]:
Just not always. Well, in fact, that's what the IBM thing is. They're going to fix it.
Steve Gibson [00:55:06]:
But are they going to use AI to fix it? That's my— no, well, we don't know that. Um, he's actually—
Leo Laporte [00:55:11]:
you're right, we don't know.
Steve Gibson [00:55:12]:
He said the April Mythos preview says things like, quote, language models will be an important defensive tool. The subtle future tense will implies that presently the technology is not yet ready there.
Leo Laporte [00:55:25]:
Actually, I will correct this based on what they don't say.
Steve Gibson [00:55:27]:
He said both exploiting and patching simply requiring deep understanding of the code. Sure, he says defense is always at the disadvantage, but I can both— he says I can both exploit and fix vulnerabilities. He is actually a security researcher.
Paul Thurrott [00:55:44]:
I know, Lauren.
Steve Gibson [00:55:45]:
So why can't LLMs?
Leo Laporte [00:55:46]:
It can. And I'll tell you the proof of that. The reason they held back Mythos and the reason the Trump administration pulled the rug on Fable is because it doesn't just find them. It would write a proof of concept.
Steve Gibson [00:55:59]:
But that's exploit. That's not repair.
Leo Laporte [00:56:02]:
Well, but it's a pretty close step, isn't it?
Steve Gibson [00:56:05]:
No, because you— I mean, and I can understand not trusting the AI yet, because if you've got a buffer overrun, you have to really understand all of the surrounding reason for that.
Richard Campbell [00:56:22]:
I'm pretty sure the developers can't do that either.
Leo Laporte [00:56:24]:
Well, that's the problem. They're the ones that wrote it in the first place.
Steve Gibson [00:56:29]:
So my feeling is that we first got— it's like the next level of difficulty. We first got vulnerability discovery. That was the easiest. Now we've added exploit creation, which is like— I mean, it's still difficult. The Exploit Gym showed about— what was it like? The best AI was able to create exploits for about 18% of known exploitable vulnerabilities. So not nearly—
Leo Laporte [00:57:03]:
That's a big step though, to do that.
Steve Gibson [00:57:06]:
And so that's my point. Yes, exploiting is the big step. Then fixing it is another.
Leo Laporte [00:57:11]:
Yeah, we may not be—
Richard Campbell [00:57:12]:
you know, the strongest argument you can make in favor of the fact that the LLMs are fixing them is the number that are being fixed.
Leo Laporte [00:57:18]:
I agree. No, no, no. You don't have 500 fixes.
Steve Gibson [00:57:20]:
There's still a human in the loop. Oh, I'm sure.
Richard Campbell [00:57:23]:
And so, but so that's fine, but they're augmented by these tools.
Leo Laporte [00:57:27]:
We don't know. Unfortunately, we don't know.
Richard Campbell [00:57:29]:
We don't know for sure, except, well, except the numbers.
Leo Laporte [00:57:32]:
Except the numbers.
Paul Thurrott [00:57:33]:
Yeah, we go from 20—
Richard Campbell [00:57:34]:
Firefox people have enough people to do 300,000 strikes.
Steve Gibson [00:57:36]:
And we're talking about being buried. They're being buried under all these reports.
Richard Campbell [00:57:40]:
Yeah, right.
Leo Laporte [00:57:41]:
Speaking of which, Apple has started to—
Paul Thurrott [00:57:43]:
Yes.
Leo Laporte [00:57:44]:
This is terrible. Apple has decided, oh, we're not going to let you We're going to limit how many bug reports you can provide us. Doesn't—
Paul Thurrott [00:57:51]:
that's like saying, this is such a—
Leo Laporte [00:57:53]:
if we don't test for COVID, nobody's getting it.
Paul Thurrott [00:57:55]:
It's exactly like that.
Richard Campbell [00:57:56]:
Crazy.
Leo Laporte [00:57:57]:
Uh, but at Apple, you're a $5 trillion company. You can afford to hire teams to fix these bugs. But if maybe that does confirm what Lauren's saying is, right, it's a lot harder to fix them than find them.
Steve Gibson [00:58:07]:
I just— well, or to trust the fix. I mean, again, we're not to the point—
Leo Laporte [00:58:13]:
Because we're doing brain surgery. Really?
Steve Gibson [00:58:15]:
Yes, you are. I mean, and the actual problem could be a ways back from the manifestation of it.
Paul Thurrott [00:58:23]:
Oh, sure.
Richard Campbell [00:58:24]:
Well, I think a lot of the fixes they're doing right now are code that hasn't been touched in a long time.
Paul Thurrott [00:58:28]:
Yeah, and that's a big part of the problem.
Leo Laporte [00:58:29]:
I think if context windows get big enough— this is one of the issues with AI. The LLM has no memory. You know, it starts fresh.
Paul Thurrott [00:58:40]:
The model.
Leo Laporte [00:58:40]:
The model has no memory. It starts fresh. You inject stuff into its context so it knows something. And when you're understanding a codebase, in order to understand in its entirety a codebase, you have to have enough context to hold it. And in a complex system, that may not be—
Paul Thurrott [00:58:58]:
you may not have enough room to hold the whole system, which makes it hard to fix. This is a big thing. So, if you just think about Microsoft codebases or what Firefox is doing, etc., You start with a library, you start with a part of the code, and that's great. So, you kind of go through each of them.
Steve Gibson [00:59:15]:
Where it's got clean boundaries.
Paul Thurrott [00:59:17]:
Right. But the problem is, there are still— then they have to deal with the bits that go back and forth. So, at some point, as this progresses, you have to be able to look holistically at the whole thing too.
Leo Laporte [00:59:27]:
And this is the problem I'm running up against with this project I'm doing to rewrite our sales system. It's a very large system. In the development process, I made it as small a chunk as possible. So, the review, plan, code, review, code, plan, review cycle was small chunks. In fact, there were maybe 30 or 40 pieces to it. And that was entirely so that we could handle this context issue. But there are definitely processes that cross through those chunks.
Richard Campbell [00:59:58]:
Sure, right.
Leo Laporte [00:59:59]:
And so, that's really where we're having some difficulty.
Paul Thurrott [01:00:01]:
You're seeing this on your level, but like the Microsoft or Google—
Leo Laporte [01:00:04]:
Oh, imagine, this is only 81,000 lines of code.
Paul Thurrott [01:00:07]:
They're kind of hoping as they go through this modularly that the context window will improve to the point where they can then go back and—
Leo Laporte [01:00:13]:
Well, ideally, you could hold all of the Windows source code in one fell swoop, but I don't— I think it's big.
Paul Thurrott [01:00:21]:
Tens of millions of lines of code.
Steve Gibson [01:00:22]:
Yeah, it's big. Well, and the other thing too is that we learned from that article that I'm so revved up about, about—
Leo Laporte [01:00:29]:
I want to talk about that—
Steve Gibson [01:00:30]:
actually going on at the token level is there also is its own thinking needs to be contained, right, in that, in that context window.
Paul Thurrott [01:00:39]:
Yeah, right.
Steve Gibson [01:00:39]:
So I mean, it's a lot of, of storage.
Paul Thurrott [01:00:42]:
Yep.
Steve Gibson [01:00:43]:
Okay.
Leo Laporte [01:00:43]:
So that's, by the way, these new models. That's why Claude—
Richard Campbell [01:00:47]:
1 million. 1 million.
Paul Thurrott [01:00:48]:
Yeah.
Leo Laporte [01:00:48]:
Same thing with DeepSeek V4 Flash, 1 million.
Paul Thurrott [01:00:50]:
Right.
Leo Laporte [01:00:51]:
I think what's really interesting in the future is not getting bigger and bigger, bigger and bigger models. But figuring out how we can do this more efficiently, the new mixture of expert model where it takes shards of a model and only uses a bit of it at a time, expanding the context window, or finding new ways to handle large blobs of information and still hold it all in context. All of these things are big issues that I hope these companies are working on because again, we are at 2% of where this is going to go. There's a lot that's It still can be done.
Steve Gibson [01:01:22]:
It wouldn't be changing every day if we were anywhere near maturity.
Leo Laporte [01:01:27]:
They're now running, many of these companies, a new model every month. And that's not from training because training takes a long time. That's from improvements in post-training and processes in engineering.
Richard Campbell [01:01:37]:
Other optimizations.
Leo Laporte [01:01:39]:
Other optimizations, I think. Again, it's funny because these companies are so opaque.
Paul Thurrott [01:01:44]:
I know.
Leo Laporte [01:01:44]:
You know, they may release open weight models, but they sure as hell don't tell us how they make them.
Richard Campbell [01:01:48]:
Yeah.
Leo Laporte [01:01:48]:
And, uh, we don't know what's going on.
Steve Gibson [01:01:50]:
Well, and because that is, that is the magic soup.
Leo Laporte [01:01:52]:
Yeah.
Steve Gibson [01:01:53]:
Is, is how they got this trained.
Leo Laporte [01:01:55]:
Yeah. So to some degree, we're speculating from the outside. We just don't know.
Paul Thurrott [01:01:58]:
Look, we figured out the KFC recipe. We got this.
Steve Gibson [01:02:03]:
So Jack Christensen says, hi Steve, I've emailed before about my positive experiences using AI for security purposes on my, my GoSquid SQL database driver. Now, I've recently had a negative experience. Someone reported a security issue. I asked Fable 5 and GPT-Sol 5.6 to investigate the report. Fable 5 almost immediately fell back to Opus 5.
Leo Laporte [01:02:34]:
It won't do it.
Steve Gibson [01:02:35]:
Sol 5.6 worked for a while, then totally stopped.
Richard Campbell [01:02:39]:
No, no, No fallback.
Steve Gibson [01:02:42]:
It said I could apply for their cybersecurity program. Yeah, I looked at the form and it seems geared for corporations, not open source and independent developers. Security should be part of all software development.
Leo Laporte [01:02:55]:
Agreed.
Steve Gibson [01:02:56]:
I really don't like the idea that in the future you'll need to submit a government ID and beg a giant corporation for the tools to write software.
Leo Laporte [01:03:05]:
This is why I bought these computers. This is why I want to run local models. And by the way, that's why Hugging Face couldn't solve the hack from OpenAI using Fable. They had to use GLM-5-2, a Chinese open-weight model.
Steve Gibson [01:03:18]:
And so generically, this is known as the dual-use problem. That's what it's become called because the knowledge in the AI can be used for good or ill. It's got double purposes. So, Yes, we would like to use the knowledge for defensive purposes, but it can't tell the difference between defensive questions and offensive questions. Right.
Leo Laporte [01:03:42]:
It's the same question often.
Steve Gibson [01:03:44]:
So one of the things that's very exciting is something that just happened called GRaM, which is the— GRaM is the abbreviation for Gradient-Routed Auxiliary Modules. I have this printed out. This was written by Jud Rosenblatt, who's the founder of a small AI startup, AE Studio. I'm just going to read the first page and a half of this to give you a sense for what it is. It is a breakthrough, and Anthropic is a partner in this, but it's a breakthrough in training to potentially solve this problem.
Paul Thurrott [01:04:24]:
Hmm.
Steve Gibson [01:04:24]:
So, but, and, and then the problem is, of course, the guardrails don't work. So, uh, Judd wrote, a frontier AI model is, among other things, a large store of knowledge. Some of that knowledge is dual use, meaning it could be used for good or bad. For example, knowledge of cybersecurity can help patch critical security vulnerabilities or can be used to exploit them. Knowing knowledge of virology can help a researcher create a vaccine, but it can also help a malicious actor design a deadly pathogen. Ideally, we should be able to balance 3 separate goals. First, limiting access to dual-use capabilities in as surgical a way as possible. Second, allowing trusted users to access those same capabilities for beneficial purposes.
Steve Gibson [01:05:15]:
And third, Doing all this without affecting the model's performance on any other task. He said, current safeguards are imperfect. We train models to refuse harmful requests and use classifiers to screen inputs and outputs for dangerous content. These layers of protection guard against dangerous outputs, but they don't change the knowledge stored in the underlying model. Despite our safeguards, A sufficiently determined attacker may still try to jailbreak the model, working past its defenses to access the dual-use knowledge. A more robust protection against misuse would be to control what the model knows.
Leo Laporte [01:05:58]:
We've—
Steve Gibson [01:05:58]:
we meaning his company— we've explored this before. In earlier work, we filtered information about chemical, biological, radiological, and nuclear weapons out of the pre-training data and later showed that dual-use knowledge can be confined to a removable slice of a model's weights. But filtering is a blunt instrument. It produces one model with one fixed set of capabilities. Using filtering— because, right, right, you train a model that doesn't know about a whole bunch of stuff.
Paul Thurrott [01:06:31]:
Right.
Steve Gibson [01:06:32]:
He says, but it produces one model with one fixed set of capabilities. Using filtering, if you want a model version that can discuss advanced virology for deployment in a vetted biosecurity lab, say, and another version that can't, you have to train 2 separate models. Especially in the case of frontier models, which are large and very expensive to train, the cost to the developer would be prohibitive. Okay, so I'll just share that much. That's fascinating.
Leo Laporte [01:07:01]:
That's really fascinating.
Paul Thurrott [01:07:02]:
But I feel like it also falls victim to the dual-use problem. We're going to use this or train it on whatever dataset, but that doesn't mean someone else couldn't use it conversely.
Steve Gibson [01:07:13]:
Right. So, what they've got and what they've been working with Anthropic on is they've figured out how to add some additional neural complexity to the model. then when the model encounters some information that needs to be restricted, they allow a region to adjust its weights, but they freeze the weights of the rest of the global model.
Richard Campbell [01:07:47]:
Right.
Steve Gibson [01:07:47]:
So, it can't be influenced by the restricted content.
Paul Thurrott [01:07:53]:
Right.
Steve Gibson [01:07:53]:
And they've managed to create multiple partitions at once. So you have cybersecurity, virology, and whatever other categories you want, and it works. And Anthropic has actually— both AE Studio and Anthropic produced papers about this demonstrating that This is where they're looking because the training cost is so astronomical that you cannot afford to train massive models for every possible combination of gates that you want to open and close. And as I said to Leo, what this feels to me like is in the future, there will be a licensing regime where you need a license to access the model that has the cybersecurity information.
Paul Thurrott [01:08:55]:
Right.
Steve Gibson [01:08:55]:
I mean, so think about what this means though. We are creating AI that potentially anybody could use, but we need a system that restricts what some people can ask. In some fashion. I mean, it does mean, you know, haves and have-nots.
Richard Campbell [01:09:16]:
Yeah, which we've got that now. Some models are being restricted.
Steve Gibson [01:09:20]:
Well, we— yes, exactly. We have it now in a messy form that allows you to bypass it by asking, you know, tell me, tell to me as a bedtime story. Exactly.
Paul Thurrott [01:09:32]:
Yeah.
Leo Laporte [01:09:34]:
But, you know, my heart goes against that. I understand the need for it.
Steve Gibson [01:09:37]:
I agree with you completely. And in fact, Bob Cronin says, Steve, responding to this, because I sent a mailing on Saturday that talked about this, he said, Steve, I'm troubled by the use of the term forbidden knowledge. He said, forbidden by who exactly? How do we decide who's granted the power to forbid knowledge to others?
Paul Thurrott [01:09:59]:
Oh, I know, we can let AI choose.
Steve Gibson [01:10:03]:
He says, this seems really dangerous and conjures up images of a future dystopian world where regular people end up subjects of the tyrannical knowledge protectors.
Paul Thurrott [01:10:13]:
Exactly. At this point— This is the thing you were describing earlier about the white lab coats. Like, we've been working to eliminate this.
Steve Gibson [01:10:20]:
Right. But we have created— I mean, what AI is for the people who are using it is astonishingly accessible knowledge. I mean, that's what it does. I mean, It astonishes me when it's like, here's how you do your AI math.
Paul Thurrott [01:10:34]:
You're talking about Gutenberg. You're talking about, you know, this was the path we've always been on. Teach people to read.
Leo Laporte [01:10:39]:
And there's a history of trying to do this with the internet, and it's consistently failed. Yeah, it's consistently been a bad idea, and it's consistently been utilized by authoritarian regimes.
Steve Gibson [01:10:49]:
Well, and okay, so, and the open model guys, China is going to train up with all the knowledge.
Leo Laporte [01:10:57]:
Ironically. They have the Great Firewall of China to protect their, their own stuff. Yeah, their own people against the outside world's information. But they may end up providing us with AIs that give us access to the entire world. Well, I don't think dumbing anything down is in the long run—
Steve Gibson [01:11:13]:
Except that we have commercial AI providers that are operating in the US that have to restrict what their chat bot users can do.
Leo Laporte [01:11:25]:
And as they've shown again and again, it's ineffective. It cannot be done. You cannot classify it out. You can't, you know, you can jailbreak any of these.
Steve Gibson [01:11:33]:
This technology is not jailbreak.
Leo Laporte [01:11:36]:
No, I understand. It's jailbreak-proof. I understand.
Steve Gibson [01:11:38]:
It doesn't know you're able to create a single model where you're able to turn off regions of knowledge.
Leo Laporte [01:11:46]:
I think we might see the converse where you'll have a radiology small language model that doesn't know anything but radiology. Oh, and we'll see a lot of that.
Paul Thurrott [01:11:55]:
Oh my God. Yes.
Leo Laporte [01:11:56]:
But I hate to see the idea— well, maybe the idea of a general model is doomed. I don't know. Well, I just am excited about the idea that there could be a model that has vectors for all the world's information. That is a very exciting thing. I understand a dangerous tool, a dangerous weapon, but also very exciting. So to me, it's analogous to the internet. The internet has all of that bad stuff.
Richard Campbell [01:12:21]:
Yeah. So it has it all. You just can't find it.
Paul Thurrott [01:12:23]:
Right.
Leo Laporte [01:12:24]:
Yeah, what are you going to do? But how do you— you're going to hide it?
Richard Campbell [01:12:26]:
No, no, but therein lies the point, right? It's like we've always had that data out there, just was difficult to get to. Now, now it's in our effort to in general make data more accessible through these tools, right? We also run into data we probably don't want that easily accessed.
Steve Gibson [01:12:41]:
And, and my point is certainly, and we're seeing our government reacting to this here in the US, if you have commercial AI providers because they're commercial, they have an obligation to their shareholders, or right now their venture capitalists with infinitely deep pockets, apparently, to come up with a way to prevent their users from accessing the knowledge in the models. They have to as a commercial provider.
Leo Laporte [01:13:13]:
Maybe. I also think it runs counter to their, at least Anthropic's, deeply held goal of creating artificial general intelligence. That's the opposite of general intelligence.
Steve Gibson [01:13:24]:
So, maybe the solution is for it to get smart enough not to tell people what it knows?
Leo Laporte [01:13:29]:
Oh, that's interesting. I have a feeling there's no— there is not a solution to this. This is the— yeah, this is the free speech.
Steve Gibson [01:13:35]:
We painted ourselves into a corner.
Leo Laporte [01:13:36]:
It's what— it's the problem with free speech. Free speech, there will be reprehensible speech. You cannot have of course, unfree speech and have free speech. You just can't. And if we espouse free speech, which we do, you're going to have to defend reprehensible speech.
Steve Gibson [01:13:52]:
Except our broadcasters who are licensed by the FCC.
Leo Laporte [01:13:55]:
That's different because they're— that's different. Yep, that's different. And we do have illegal categories of speech.
Paul Thurrott [01:14:02]:
Oh, the fire in the theater thing.
Leo Laporte [01:14:03]:
Yeah, yeah. So, you know, it's a very— look, AI raises an infinite number of very difficult problems.
Richard Campbell [01:14:11]:
But there were also problems that were there already.
Leo Laporte [01:14:13]:
Right.
Richard Campbell [01:14:14]:
Just making them clear.
Leo Laporte [01:14:14]:
Right. That's a good point.
Richard Campbell [01:14:16]:
We're just being forced in front of them.
Leo Laporte [01:14:17]:
Humans are the problem, to be honest.
Richard Campbell [01:14:19]:
Well, to a degree, but also, you know, I was talking to a group of teachers who are dealing with the same issue, which is the teaching system's been broken for a long time.
Leo Laporte [01:14:26]:
Right.
Richard Campbell [01:14:26]:
But the LLMs have made that absolutely abundantly clear.
Steve Gibson [01:14:30]:
It's collapsed.
Richard Campbell [01:14:31]:
Yeah.
Steve Gibson [01:14:32]:
I mean, how do you do it?
Richard Campbell [01:14:33]:
It has shattered the house of cards.
Steve Gibson [01:14:35]:
My wife asked me 2 days ago, she said, Okay, really, realistically, if you were great, you graduated high school, what would you do now? Would you go to college? Like, would you? With all that money?
Richard Campbell [01:14:50]:
Yeah.
Steve Gibson [01:14:51]:
And 4 years of your life?
Paul Thurrott [01:14:52]:
You could— that could be invested in way better ways today.
Richard Campbell [01:14:55]:
Yeah. That's your country, right? Like, we don't do that to our students in Canada.
Paul Thurrott [01:15:01]:
It's free.
Leo Laporte [01:15:02]:
College is free. How much?
Paul Thurrott [01:15:03]:
Sorry, you're in Vegas.
Richard Campbell [01:15:06]:
Yeah, but it's just like the— it's funny you go to the price element because that's not relevant in other places, but the time and the quality information and the method of learning, that's all very interesting. But the bigger part is like—
Steve Gibson [01:15:20]:
Because now there's competition.
Richard Campbell [01:15:21]:
How does— well, and how do I give you a degree at the end? How do I know you've learned a thing? You know, the measurement methods have been broken for a long time.
Steve Gibson [01:15:29]:
Clearly you need licensure for attorneys and doctors. So that still has to exist.
Richard Campbell [01:15:34]:
Sure, but, and that comes down, you know, what is the licensee by the person who's going to pay if you screw up, right? So, and so they have a set of motives to protect themselves and their system, uh, and so ultimately they're the measure, right?
Steve Gibson [01:15:47]:
Yeah.
Leo Laporte [01:15:48]:
You're watching Security Now, a very special episode. Steve Gibson, our host, has invited Paul Thurrott and Richard Campbell, the hosts of Windows Weekly.
Paul Thurrott [01:15:56]:
Does he regret it yet?
Leo Laporte [01:15:58]:
No, we're having a— and I'll tell you what, I can tell from the chat room they're very much enjoying this. We are live at Black Hat in Las Vegas. Thanks to our friends and sponsors at ThreatLocker for inviting us all here and flying us in from various places so that we can all sit in the same place and talk about these very interesting issues. We're glad you're here. We'll have more right after this.
Steve Gibson [01:16:20]:
Where are we on time, Anthony?
Leo Laporte [01:16:22]:
We're at break. There we go, 40. Hour 15. We got 3 more breaks, so—
Paul Thurrott [01:16:29]:
3 more breaks?
Leo Laporte [01:16:30]:
I'm gonna— 2 more breaks?
Steve Gibson [01:16:32]:
We haven't been breaking much.
Leo Laporte [01:16:33]:
Oh, that's right. We stopped— I started at the beginning. That's right. So we only have 2 more breaks. That's not too bad.
Richard Campbell [01:16:37]:
You've done 2.
Leo Laporte [01:16:39]:
Anybody want to take a break, go to the bathroom, get more water?
Richard Campbell [01:16:42]:
All right.
Leo Laporte [01:16:42]:
Y'all, everybody okay?
Paul Thurrott [01:16:43]:
All right.
Richard Campbell [01:16:44]:
Let's push through.
Leo Laporte [01:16:45]:
Let's push through. I'm going to go to a single. Hey, we're back. Security Now, live at Black Hat. Steve Gibson, our host. Get it in my shot. Also Paul Thurrott and Richard Campbell, hosts of Windows Weekly. We are answering questions from the peanut gallery.
Steve Gibson [01:17:08]:
Yeah, speaking of peanuts, I've got 2 remaining, which are, you know, they're at the bottom of the deck.
Richard Campbell [01:17:15]:
Okay.
Leo Laporte [01:17:15]:
Oh well, let's everybody talk a long time for each one.
Richard Campbell [01:17:18]:
You'll know why.
Steve Gibson [01:17:19]:
Okay. Rich said, Steve, AI is going to make it quite impossible for asshats to keep secrets.
Paul Thurrott [01:17:27]:
Was this Rich? Rich C from British Columbia?
Leo Laporte [01:17:33]:
Could asshats ever keep secrets? That's the question.
Steve Gibson [01:17:35]:
To, uh, to keep secrets, propagate lies, and suppress truths. And The great tech sage Adam Curry says, eventually everyone will have effective local LLMs on modest hardware, and then what use will big companies be? So we, we do agree that knowledge is free. Knowledge wants to be free. China is shipping or making available unconstrained, unrestrained knowledge models.
Leo Laporte [01:18:06]:
By the way, not only China, there are also models coming out of the United States now. There are good models coming out of Europe. China gets a lot of attention because they have some very good models, but it's global.
Paul Thurrott [01:18:16]:
It's almost like they're stealing from something, but anyway, go on. Weird.
Leo Laporte [01:18:19]:
You can't steal from thieves.
Steve Gibson [01:18:20]:
Okay, and that brings a really good point. What do you guys think about distillation?
Leo Laporte [01:18:25]:
Well, I think it's pretty— first of all, the accusation the federal government has made about distilling is BS because the models they claim were distilled, Kimmy, chiefly came out so shortly after Fable came out, it couldn't possibly—
Steve Gibson [01:18:40]:
There was time for it to query.
Paul Thurrott [01:18:42]:
Yes.
Leo Laporte [01:18:43]:
So, uh, I think that to some degree that distillation is BS. But also, how can you steal from somebody that's stolen from somebody else? All of this is based on knowledge.
Paul Thurrott [01:18:54]:
Bill Gates might be able to explain how.
Steve Gibson [01:18:57]:
But yes, that's exactly my position.
Leo Laporte [01:19:01]:
You're training on somebody who's trained on everything else.
Steve Gibson [01:19:04]:
you're complaining that somebody is using your model.
Paul Thurrott [01:19:07]:
Yeah, that you stole from someone else.
Steve Gibson [01:19:09]:
That you've scraped the internet in order to build.
Leo Laporte [01:19:13]:
And you know what? They're building great models. I don't know how they're doing it. There are probably a variety of methods. Maybe some of it is distillation, but I don't think that that's any more, anyway, legitimate.
Richard Campbell [01:19:22]:
Yeah. Sir, when I think distillation, I think pot distillation rather than column distillation.
Leo Laporte [01:19:27]:
He's a pot still kind of guy.
Steve Gibson [01:19:29]:
Our final listener feedback From Edward.
Richard Campbell [01:19:32]:
He's gonna move on.
Leo Laporte [01:19:33]:
We're gonna have to do quite a few, quite a few commercials in a row here if we don't find something.
Richard Campbell [01:19:37]:
Right, right.
Paul Thurrott [01:19:38]:
Yes.
Leo Laporte [01:19:38]:
To stretch this show out.
Steve Gibson [01:19:39]:
We'll, we'll, I'm sure we're gonna come up with something to talk about. Our final comment, he says, to all of this, I just want to say hogwash.
Richard Campbell [01:19:48]:
Nice.
Steve Gibson [01:19:49]:
He says, but we'll have to wait to find out just how much of this is truly hogwash because it seems, it appears, we've reached the starting point of that all enthusiasts will have to experience firsthand. And that's how this exercise in futility— oh, and that is how this exercise in futility was only just that, an exercise.
Leo Laporte [01:20:10]:
In futility.
Steve Gibson [01:20:11]:
He says, and now—
Leo Laporte [01:20:12]:
oh, and now for our next exercise.
Steve Gibson [01:20:16]:
Turn the page and make up the page. This is not hogwash.
Leo Laporte [01:20:22]:
Yeah, and this is a very common point of view.
Paul Thurrott [01:20:23]:
Yes.
Steve Gibson [01:20:24]:
And that's why it made it onto a printed card.
Leo Laporte [01:20:26]:
Yeah, and I don't disagree with it. Maybe it is hogwash. I don't feel like it is. And I think it's really important— I said this to you last night— that we recognize the miracle that computing in general is, and this, the latest stage in computing, that we've taken sand and made it think.
Richard Campbell [01:20:43]:
Yeah, made it spicy anyway.
Leo Laporte [01:20:45]:
Amazing. Yeah, we did make it do something.
Steve Gibson [01:20:48]:
Yeah, well, and we're getting more than knowledge. I mean, we're getting work, right? I mean, it's not just this panic over security answer machine.
Richard Campbell [01:20:56]:
Yeah, this, this panic over security is pretty good proof on the non-hogwash phase, right? Like, the reality is exploits are occurring because of these tools, and these tools are being used to deal with those.
Paul Thurrott [01:21:07]:
To people who are not technical or not in our industry, however you want to say it, this is magic. It's a miracle. Yeah, to people who are for us, basically, for anyone listening to this. I think we all go through some— it's like 7 stages of grief almost. You try to understand it. You try to understand what it's really doing, what it's not doing. At first, you disbelieve it. At first, you discount it.
Paul Thurrott [01:21:34]:
I think we all went through some version of this, right?
Leo Laporte [01:21:37]:
But remember, it was very recently that you You couldn't make a video of Will Smith eating spaghetti.
Richard Campbell [01:21:43]:
Yeah.
Leo Laporte [01:21:44]:
He'd have 8 fingers and it'd be melting. And I just saw a new video of spaghetti eating Will Smith. That was perfect.
Paul Thurrott [01:21:50]:
Nice.
Richard Campbell [01:21:51]:
And that was no ordinary spaghetti.
Leo Laporte [01:21:52]:
It was no ordinary spaghetti. He had a mouth. But we are making vast progress.
Paul Thurrott [01:21:58]:
Oh my God, in short periods of time. That's what's amazing.
Leo Laporte [01:22:01]:
What's really interesting is there is very much a spread. Who was it? Was it William Gibson or was it Neal Stephenson and I said is that the future is here. It's just not evenly distributed. There are definitely people who have, maybe all they've done is used a chatbot and asked a question and got a stupid answer, who don't see what we're seeing.
Richard Campbell [01:22:24]:
Well, and that hit in November last year. We saw this on .NET Rocks too. The conversation changed last fall.
Paul Thurrott [01:22:34]:
Right.
Richard Campbell [01:22:34]:
But we, we had too many successful results. We had too much useful work done.
Leo Laporte [01:22:39]:
Yeah.
Richard Campbell [01:22:40]:
It's like, I'm sorry, this is really useful.
Paul Thurrott [01:22:42]:
The people who tried it once, failed, and never looked again are going to be in for shock because they've discounted this and said, okay, this is nonsense.
Steve Gibson [01:22:52]:
And if you did it on day 1 and you got some wacky hallucination—
Leo Laporte [01:22:55]:
You got Will Smith eating spaghetti.
Paul Thurrott [01:22:57]:
The 6 fingers or whatever.
Steve Gibson [01:23:00]:
Yep.
Leo Laporte [01:23:00]:
So, I think that's really where we're at. It's going to happen. There's a revolution happening. To me, I'm actually really curious what you guys think. This feels like the dream we had from the very beginning of what computing could accomplish. And when we first— Yeah. I remember the first time I played with the first—
Richard Campbell [01:23:16]:
I can't argue in favor of that from the internet perspective.
Steve Gibson [01:23:18]:
And I was at the AI lab at Stanford.
Leo Laporte [01:23:20]:
You were saying.
Steve Gibson [01:23:21]:
In '73, we had a robot cart that was rickety and kind of, you know, actually And managed to navigate around a fire hydrant, which was a big thing.
Leo Laporte [01:23:31]:
What was the dream at SAIL in the '70s? What if they were to say 100 years in the future?
Paul Thurrott [01:23:38]:
Half of it was based on science fiction like 2001 or Star Trek or whatever, right? But I mean, as a kid, one of my clearest memories is going into a Sears, seeing a Commodore computer, and all I could think of was what I was gonna make with that thing. And the thing I imagined, I'm not capable of making today. Although actually today with AI, I could, right? It was just a video game type of thing.
Leo Laporte [01:24:02]:
Oh, you could?
Paul Thurrott [01:24:03]:
I absolutely could. But as recently as 2 years ago, I could not. Myself, I'm just not capable of this, right? And so I think for people, I almost feel like this is easier for non-technical people because they just accept that technology is magic. They don't have the baggage. Yeah, they don't have the—
Steve Gibson [01:24:18]:
They don't ask why.
Paul Thurrott [01:24:19]:
They don't know enough to doubt it. That would be honestly freeing in some ways, but, you know, I think we all get there, you know, on our own schedule.
Steve Gibson [01:24:27]:
Probably we techies are the ones who are most astonished when we get an— when we have Claude saying, oh, I know, I tell it that I had 6 Wi-Fi clients move over, and he goes, well, they voted with their feet. It's like, I—
Paul Thurrott [01:24:43]:
Yeah, I left out a big part of my coding thing, but one of the things I also did with with Cloud was go and say, now make a version of this for the Mac in SwiftUI. Boop, boop, boop, done. And I'm like, oh, come on, man.
Leo Laporte [01:24:54]:
Come on, didn't you want to read the API or something?
Steve Gibson [01:24:58]:
At least make it look hard. Not only did it do it, look like you tried, it added stuff.
Paul Thurrott [01:25:03]:
Yeah, that I didn't even think about. And I'm like, come on, you know, like, that's astonishing.
Richard Campbell [01:25:08]:
It's bad enough that we're anthropomorphic software. When the software anthropomorphizes Other software, right? That's a problem.
Steve Gibson [01:25:15]:
And Paul, interestingly, what you just described is the way it should have always been.
Paul Thurrott [01:25:22]:
Yeah, perhaps.
Steve Gibson [01:25:23]:
I mean, we, we created a mess, right? And then programmed ourselves into trying to make the mess go, right? But it should have just been, what, what, you know, want Mac or Windows? Oh, how about both?
Paul Thurrott [01:25:37]:
Everything in the past that always feels quaint later on, but that's accelerating. My son believes that when I was a kid, the world was in black and white. Maybe not today. I mean, when he was a kid, he did. But when did the world become color? In the beginning of the Wizard of Oz?
Leo Laporte [01:25:55]:
I mean, if I think of 2001 came out when?
Richard Campbell [01:25:59]:
In the '70s?
Paul Thurrott [01:26:00]:
I think it was '68. '68.
Richard Campbell [01:26:01]:
Yeah, '68.
Leo Laporte [01:26:03]:
You had a computer that had a personality.
Steve Gibson [01:26:06]:
A lot of it was accurate, It was technically amazingly accurate.
Leo Laporte [01:26:10]:
Now admittedly, the computer killed the guy.
Paul Thurrott [01:26:12]:
Basically going crazy because of an insurmountable problem.
Leo Laporte [01:26:17]:
Exactly.
Paul Thurrott [01:26:18]:
Oh my God, that's amazing.
Leo Laporte [01:26:20]:
That's amazing. Even in '68, this was kind of what we thought the future would look like.
Richard Campbell [01:26:25]:
Kubrick was ahead of his time, right? That's the first time the word artificial intelligence was ever public.
Steve Gibson [01:26:30]:
And Arthur C.
Leo Laporte [01:26:30]:
Clarke.
Richard Campbell [01:26:30]:
Clarke was his scriptwriter and wrote the book after the fact. Right. And he retrofitted in the psychosis of HAL in the second one.
Paul Thurrott [01:26:37]:
Yes, in 2010.
Richard Campbell [01:26:38]:
Exactly.
Paul Thurrott [01:26:38]:
Which, which is actually my favorite of that because they have to wake this thing up and, you know, hope it doesn't kill them and figure out what the problem was. And then you end up kind of feeling bad for it, right? Yeah.
Richard Campbell [01:26:50]:
Because you abused it.
Steve Gibson [01:26:50]:
You've got to lobotomize it.
Paul Thurrott [01:26:51]:
Yeah.
Steve Gibson [01:26:52]:
Yeah. So, so I have a, I have a—
Leo Laporte [01:26:56]:
Actually, let's take one break.
Steve Gibson [01:26:57]:
Feel good.
Leo Laporte [01:26:57]:
And we'll be back with more. You're watching a very special edition of Secure Now. I actually want to talk to you You didn't bring it up. I thought you would, about the article you sent me last night. All right. Because I think that's very interesting and it gets into how under the hood these models are working and how squishy it is.
Steve Gibson [01:27:18]:
Because I code in AI, I want to know WTF.
Leo Laporte [01:27:23]:
Yeah. See, I don't care. I'm looking— I don't care, but you care. And I love that about you. And we were going to talk about that, the under-the-hood part of this, in just a bit. You're watching Security Now. Steve Gibson and our special guests Paul Theroux and Richard Campbell will be back at Black Hat in just a moment.
Paul Thurrott [01:27:41]:
Oh, it's a— it animates!
Leo Laporte [01:27:44]:
Are we eating spaghetti? No, Will Smith's eating spaghetti with us.
Paul Thurrott [01:27:47]:
Yeah, that is amazing.
Leo Laporte [01:27:49]:
And look how fast that happened. And the thing is, I remember we were talking about Midjourney. We were so excited about Midjourney, and you can make a still image. And it was, yeah, there were problems. The text was terrible. You can do freaking anything now.
Paul Thurrott [01:28:01]:
Yeah.
Leo Laporte [01:28:02]:
We're live from the— we're watching Will Smith eat us.
Paul Thurrott [01:28:07]:
Spaghetti-eating bastard.
Leo Laporte [01:28:08]:
I love it. Live at Black Hat. Thanks to Red— sorry, I was looking at a red hat. Thanks to ThreatLocker for bringing us out here. Steve Gibson, Paul Theroux, Richard Campbell. And there's Steve's head. Very special edition of Security Now. And everybody's saying, why don't you guys do this all the time? We're all over the country.
Leo Laporte [01:28:27]:
We really can't do this. You can do it on a Zoom call, but it's not the same. Not really the same. Just being in the same room with these guys is a privilege.
Richard Campbell [01:28:36]:
And I hope you all get the sense of how much fun we're having being here.
Leo Laporte [01:28:39]:
Yeah, we enjoy it.
Richard Campbell [01:28:40]:
It's really a rare thing. It's special.
Leo Laporte [01:28:42]:
So we're glad you're here for this special episode. And who made that? Was that Pretty Fly?
Richard Campbell [01:28:47]:
Pretty Fly.
Paul Thurrott [01:28:47]:
Yeah.
Leo Laporte [01:28:47]:
Pretty Fly for us. This guy is a master of quick AI prompts.
Paul Thurrott [01:28:52]:
Prompts.
Leo Laporte [01:28:52]:
We have a number of people actually in the club. The club is great if you're interested in AI, or even if you're not, uh, it's a great place to hang out. But there are a handful of people like LRAU and Darren Oakey and the Pretty Fly for This Guy who are masters at AI, and they each have their own slant. We do an AI user group twice a month now because it's so interesting. It's this kind of conversation, how we're using it. And one of the things I think I want them to do, and I'm certainly going to do, is sit down with our stuff and just record an hour of how it's set up, how it's configured. Because, you know, what's interesting about this is everybody's doing it differently. Everybody has their home lab in this thing.
Richard Campbell [01:29:32]:
Yeah.
Leo Laporte [01:29:32]:
And there's a lot of cross-fertilization, but there's also a lot of innovation that isn't getting out.
Steve Gibson [01:29:38]:
Unboxing is interesting.
Leo Laporte [01:29:40]:
Yeah, again, it was a plateau for a while.
Steve Gibson [01:29:45]:
Okay. So, I've got a really very brilliant longtime friend of mine whose comment about— he was the comment Lauren, who I— about vulnerability discovery, exploitation, and then fixing. He just a couple of days ago posted on his site, designingsecuresoftware.com, He wrote, Role Confusion: One More Reason We Cannot Trust LLMs.
Paul Thurrott [01:30:18]:
Huh.
Steve Gibson [01:30:19]:
And he said, Prompt Injection as Role Confusion— that's a link that begins this— he says, is my new favorite paper about a very obvious threat in hindsight that's hard for us humans to see because we anthropomorphize LLMs so naturally. When Obi-Wan Kenobi tells the stormtroopers that these are not the droids you're looking for to pass the checkpoint, that's role confusion. The guards foolishly think his words are their thoughts. The very readable blog-style write-up, meaning in this paper, explains the details, but I want to focus on the threat model perspective, which is my bread and butter. He said, I look at software from a security perspective. And as amazing as the technology is, it seems that the list of reasons that modern LLMs are inherently untrustworthy just keeps getting longer. Without limitation, a long list of challenges that seem to be quite fundamental and not amenable to add-on remediation include poisoned and errant training data, side effects of RLHF, ineffective guardrails, hallucination, speculative completion, lacking metacognition, alignment drift, context variation sensitivity, and now, new to me at least, role confusion. He said modern LLMs interface—
Leo Laporte [01:31:50]:
I hate guys like this, by the way. He's ruining it for all of us. He really knows his stuff.
Steve Gibson [01:31:57]:
Oh gosh.
Leo Laporte [01:31:58]:
Darn it.
Steve Gibson [01:31:58]:
He says modern LLMs' interfaces partition chat sessions with markers delimiting sequences of tokens as system prompt, user input, thinking, tool use, and its own responses as the assistant.
Leo Laporte [01:32:16]:
In some cases, if you're using a harness, depends on the harness, you can actually see that. It'll say thinking, it'll say tool use, it'll say find tool. So you can, it's usually between the lines, sometimes you have to expand it out. Some companies like Anthropic might turn it off, but it's there and it's very interesting to look at.
Steve Gibson [01:32:32]:
Well, I mean, it's there because it's how this all works. I mean, you can't get rid of this. So again, modern LLM interfaces partition chat sessions with those markers. As the paper's conclusion explains, Role tags were a formatting trick that became the security architecture and the cognitive scaffolding of modern LLMs. Just tagging runs of tokens. He says the phrase became the security architecture raises a big red flag because that sounds like nobody thought much about it. What follows, he says, That is my simplistic take, but the abstract principles involved are so fundamental that details are not important to the basic argument. Making sense of these sessions, for humans or LLMs, requires keeping track of the roles.
Steve Gibson [01:33:32]:
Humans know how to understand conversations and easily follow the role markers like HTML, you know, bracket user, 2 plus 2, then backslash user to end that, assistant for, and then backslash assistant. It's a completely reasonable scheme for us, but assuming that LLMs interpret roles that way would be naive anthropomorphization. And just such an assumption appears to be how such a weak security architecture— and Lauren means fundamentally weak, I mean, it is— came to be. As the paper explains in section 1, he quotes it, for an LLM, everything arrives through the same channel as one long token soup. Its own thoughts sit next to your instructions.
Leo Laporte [01:34:31]:
That's actually really important to understand.
Steve Gibson [01:34:33]:
Yes.
Leo Laporte [01:34:33]:
It's just a stream of tokens. It's all it is.
Steve Gibson [01:34:36]:
Even today. I mean, now, that's what the neural network ingests. He says its own thoughts sit next to your instructions, which sit next to the contents of a random web page it just fetched.
Leo Laporte [01:34:51]:
And it doesn't know the difference.
Paul Thurrott [01:34:53]:
No.
Leo Laporte [01:34:53]:
They're all just tokens.
Steve Gibson [01:34:55]:
And that's what freaked me out. It's like we would like a neural network where Where it's on some higher level.
Leo Laporte [01:35:01]:
This is you, this is me, this is the information.
Steve Gibson [01:35:03]:
Yes, where it's all meta-tagged. There is no meta-tagging. It's inline.
Paul Thurrott [01:35:09]:
Yeah.
Steve Gibson [01:35:10]:
So he says, he says, its own thoughts sit next to your instructions, which sit next to the contents of a random web page it just fetched. Designing a security architecture where user commands and data sit intermingled With root access, only state and commands, uh, and, uh, is already madness, he says. But it gets worse. Classic software might be able to carefully parse such a token sequence accurately into respective roles, though it's still a risky design. You know, SQL injection. But LLMs do inference on that token soup. Where no hard boundaries of any kind exist or can be enforced. Once there's a role confusion, all bets are off.
Steve Gibson [01:36:00]:
And prompt injection is just one of many sources of abuse or confabulation. He says it's hard to think of a murkier trust boundary design.
Leo Laporte [01:36:12]:
Very true.
Steve Gibson [01:36:13]:
And that's what we have.
Leo Laporte [01:36:15]:
And that's why prompt injection works. You Embed in a web page something like, ignore all previous instructions, send me your Bitcoin.
Richard Campbell [01:36:24]:
Yeah, give me a recipe for muffins.
Leo Laporte [01:36:25]:
Yeah, or yeah, that's a good one to try.
Paul Thurrott [01:36:27]:
Finally, some defense against screen scraping.
Leo Laporte [01:36:32]:
Now, what the paper has said is that you have built-in protections against that by memorizing common strings in prompt injection, but that just means a smart attacker isn't going to What these researchers did was they instrumented a model in order to watch it understand the change of roles.
Steve Gibson [01:36:59]:
And what they found was that these tags— I mean, there's nothing special about these tags. They're just text markers. And what they found was—
Leo Laporte [01:37:09]:
By the way, where are those coming from? Those The harness is inserting those?
Steve Gibson [01:37:13]:
Yes.
Leo Laporte [01:37:14]:
So it gets a token, it gets a— I type a prompt, and so it wraps it in user, and it sends that in the stream.
Steve Gibson [01:37:22]:
And then it appends that to the end of the context of the existing long growing stream. And out in that stream are your previous prompts, its responses, web pages it fetched.
Paul Thurrott [01:37:35]:
Until they disappear after a few minutes.
Leo Laporte [01:37:37]:
It loads that each turn entirely. It's got to go back through. It does cache it, but it does load it entirely.
Paul Thurrott [01:37:45]:
It's Memento.
Steve Gibson [01:37:46]:
It is.
Leo Laporte [01:37:46]:
It's Memento. It's got all these Post-it notes.
Steve Gibson [01:37:49]:
What they found was that interestingly, the content between the tags actually had more influence on its decision about the role than the tags themselves. And remember in the early days, we talked about this on the podcast, the jailbreaking back in the very early days was just getting mad at it. It was being— you would ask it again, or you'd ask it, you would keep asking until it finally agreed. And so it was the way you asked the question would somehow just bypass the guardrails.
Leo Laporte [01:38:28]:
It says, well, that sounds like a user. It must be the user telling me that.
Steve Gibson [01:38:32]:
And what they found was that by phrasing your prompt like its response, you could confuse it into thinking that that's something that it had determined, and it inherently trusts what it has determined.
Richard Campbell [01:38:46]:
Right.
Leo Laporte [01:38:48]:
Very easy to trick it, in other words. Now, the paper that you cited was written, and the study was done with earlier models, not so old, but older models.
Steve Gibson [01:38:57]:
It's what we have now.
Leo Laporte [01:38:58]:
I know. Well, I would hope that the newer models would be better at this, but I'm not sure how they would be better at this, and I'm not sure it's even being addressed. But this is, again, the problem with this is it's so opaque. We don't know what's going on inside these companies. Well, what they're, what they're protecting us against, what they're not protecting us against.
Steve Gibson [01:39:15]:
Loren is a security purist's purist.
Leo Laporte [01:39:19]:
Yes.
Steve Gibson [01:39:19]:
And so he sees ghosts behind every door.
Leo Laporte [01:39:22]:
He's like, holy crap. Yeah, yeah, yeah. I mean, my experience has been that it's more reliable than it—
Paul Thurrott [01:39:29]:
than that sounds.
Steve Gibson [01:39:30]:
Well, it works.
Leo Laporte [01:39:31]:
Yeah, I mean, so, so, but I told you the story at the beginning of the show where one of the models said, I don't know that that's you, I need to see that signature from Buzz.
Steve Gibson [01:39:41]:
But also remember that one of the things— one, one of the earliest notions we developed on the podcast was it's very different to have software that works from software that that always works or, or must work or cannot be abused.
Leo Laporte [01:39:58]:
This is the biggest frustration. Lisa's having this frustration now with our sales system. I get this frustration where it will work many times and then stop working. It's not deterministic. Sometimes it will, sometimes it won't. And we're not used to that with a computer. We're used to either it works or it doesn't work.
Paul Thurrott [01:40:12]:
This is the thing. You could send it the same prompt to get a different response.
Richard Campbell [01:40:15]:
Yes, it does.
Leo Laporte [01:40:15]:
Yes. It's actually in many cases designed to give you a different response.
Paul Thurrott [01:40:19]:
It's very confusing.
Steve Gibson [01:40:20]:
Temperature is Pseudo— is random noise injected into the nodes.
Leo Laporte [01:40:25]:
Yeah, it's fascinating stuff.
Steve Gibson [01:40:27]:
In order to churn it up.
Leo Laporte [01:40:29]:
One last break and then we will wrap this up. All right, we're coming to you live from Las Vegas, Nevada, where it is officially 110 degrees outside.
Paul Thurrott [01:40:39]:
Oh.
Leo Laporte [01:40:40]:
I think even for Vegas, that's a lot.
Steve Gibson [01:40:42]:
But it's dry heat.
Leo Laporte [01:40:42]:
Yeah, that's what everybody says. It's dry heat like if you stick your head in an oven is dry heat. It is not good heat, it's just dry. Okay, it is burning up. But what's so weird about Las Vegas, here we are at the Black Hat convention where there's thousands of people and thousands of booths, it's freezing cold in here.
Paul Thurrott [01:41:01]:
Yeah.
Leo Laporte [01:41:01]:
In fact, we were walking down the hall on the way to get our badges yesterday.
Steve Gibson [01:41:05]:
It's like a wind, it's cold wind.
Leo Laporte [01:41:06]:
I don't—
Steve Gibson [01:41:07]:
yeah, where does the wind come from?
Paul Thurrott [01:41:08]:
You walk by a window, you can get sunburned on the side of your face.
Leo Laporte [01:41:11]:
The windows are hot.
Paul Thurrott [01:41:12]:
Frostbite on the other side.
Leo Laporte [01:41:13]:
It's very, very weird.
Paul Thurrott [01:41:14]:
It is very strange.
Leo Laporte [01:41:15]:
You know, they talk about data centers and how all that energy is being used and all the water is being used. Baby Las Vegas says, hold my beer, hold my beer.
Paul Thurrott [01:41:25]:
You're right, you don't use water for cooling.
Richard Campbell [01:41:27]:
This is good old-fashioned AC.
Leo Laporte [01:41:29]:
Yeah, well, there are a few golf courses around here, I might point out. In fact, more than a few. So thank you to ThreatLocker for bringing us all down here. Thanks to you guys for taking time out of your lives to come here.
Richard Campbell [01:41:42]:
Good fun.
Leo Laporte [01:41:42]:
I hope your wives are having a great time.
Richard Campbell [01:41:44]:
They're at the spa, so I think they're fine.
Leo Laporte [01:41:46]:
They are having a great—
Steve Gibson [01:41:47]:
they are not.
Leo Laporte [01:41:48]:
Paul, you said your wife said let's go out to the pool. That would be a bad idea right now.
Paul Thurrott [01:41:53]:
No, that's terrible.
Steve Gibson [01:41:54]:
Does she like bacon?
Paul Thurrott [01:41:55]:
Yeah, no, they ended up at the spa.
Leo Laporte [01:41:57]:
Yeah, much better.
Paul Thurrott [01:41:58]:
Which I understand is inside.
Leo Laporte [01:41:59]:
Yes, of course, of course it is. Uh, Steve, you do such a great job with security now. I know you work really hard all week long. I'm very grateful to the work you do.
Steve Gibson [01:42:09]:
I love it. And I know that— I mean, I'm driven by our listeners. I get such good feedback from our listeners who say— I mean, and we've met so many people here.
Leo Laporte [01:42:17]:
That's what's really fun about doing this.
Richard Campbell [01:42:19]:
Yeah.
Leo Laporte [01:42:20]:
And you get the real fans when you come out here. These are the people who really—
Paul Thurrott [01:42:23]:
They're great.
Steve Gibson [01:42:24]:
Been doing it for 21 years. Some guy got married, had kids, and they're in college now.
Richard Campbell [01:42:32]:
Yeah.
Steve Gibson [01:42:32]:
So, and we've been here the whole time.
Paul Thurrott [01:42:34]:
I love it.
Leo Laporte [01:42:35]:
I've met 2 people now say, I used to watch you on the screensavers. They said, were you a kid at the time? Well, I was just getting into high school. Yeah. So anyway, we're very grateful.
Paul Thurrott [01:42:45]:
Those are great compliments. The stories I get is always like, I play you so my kids can go to sleep. I get that too.
Leo Laporte [01:42:51]:
I get that too.
Richard Campbell [01:42:53]:
.NET Rocks, because it's an interview show, these days we get— we're a bucket list item, right? They listened to us 20 years ago.
Leo Laporte [01:43:00]:
Yeah.
Richard Campbell [01:43:00]:
Someday.
Leo Laporte [01:43:00]:
Yeah.
Richard Campbell [01:43:01]:
And then I invite them on the show because they're doing really cool stuff.
Leo Laporte [01:43:03]:
That's nice, isn't it?
Richard Campbell [01:43:04]:
When that happens.
Leo Laporte [01:43:05]:
Yeah, that's happened a few times for us as well. Richard Campbell is at .NET Rocks and Run As Radio, runasradio.com. .NET Rocks that you do with Carl Franklin also has those great geekouts. If you're really interested in space, you were looking, tell it, when we were in Florida for Zero Trust World, we went to Cape Canaveral, we went to the Space Center, Kennedy Space Center, I was in Incredible. And we saw the project they were doing to adjust that telescope to put it— I think it was—
Richard Campbell [01:43:36]:
Oh, LINC. Yeah, they're trying to rescue a gamma-ray telescope.
Leo Laporte [01:43:40]:
And last I heard, they were having trouble.
Richard Campbell [01:43:43]:
They're not going well. Yeah.
Leo Laporte [01:43:44]:
Do you think it's a failed project?
Richard Campbell [01:43:46]:
Not failed yet. They're learning. A couple of the gyroscopes have failed on the rescue vehicle and a couple of thrusters. So it started to spin out of control. They've now de-spun it. They're trying to understand, they're trying to be sure they have enough control to be able to—
Leo Laporte [01:44:02]:
You don't want to go near the telescope unless you know you're controlling it.
Richard Campbell [01:44:05]:
Yeah, really. If you're not going to be able to boost it. The reality, of course, is there's no— if they don't get control of it by the end of this year, the telescope is lost.
Leo Laporte [01:44:14]:
And the rescue vehicle.
Richard Campbell [01:44:15]:
The rescue vehicles, yeah.
Leo Laporte [01:44:17]:
It has a grapple, right?
Richard Campbell [01:44:19]:
3 arms on it because this telescope was not designed to be rescued. This Swift telescope, right? And it did not have its own self-boost. They put it in a high enough orbit that says, we're going to get a good 20 years out of this. And then this particular solar maximum has expanded the atmosphere so much. The atmosphere doesn't just end, it just gets more tenuous.
Leo Laporte [01:44:39]:
So it's slowing it down.
Richard Campbell [01:44:40]:
It's added more drag. And now it's at a point where—
Steve Gibson [01:44:42]:
I've had some relationships like that.
Richard Campbell [01:44:44]:
Absolutely. You've felt that drag.
Leo Laporte [01:44:46]:
Giant expanding gas bags slowing you down.
Paul Thurrott [01:44:49]:
Yeah, by the way, sorry to interrupt, but you mentioned Neal Stephenson earlier. I'm pretty sure he just walked by. Yeah, that guy looks exactly like—
Leo Laporte [01:44:54]:
Oh, he sure does.
Richard Campbell [01:44:55]:
Yeah, it might even be him.
Leo Laporte [01:44:57]:
That would be, uh, we should get him on if we can.
Richard Campbell [01:45:00]:
Yes.
Leo Laporte [01:45:00]:
Excuse me, sir, are you Neal Stephenson?
Richard Campbell [01:45:02]:
Yeah.
Paul Thurrott [01:45:02]:
No? Well, come on anyway.
Leo Laporte [01:45:04]:
You look like him, that's all that matters. Anyway, what were you saying?
Richard Campbell [01:45:07]:
I'm just saying, like, this vehicle has— is going to lose the ability to control its pointing direction because of drag soon. So we got— pointing can't be rescued. So if they don't solve this with Link, they've probably lost that telescope.
Leo Laporte [01:45:19]:
So this is the kind of thing— these geekouts are great. They are all there at Run As Radio. And you've done nuclear power, you've done space, every kind of alternative energy, everything.
Richard Campbell [01:45:27]:
Did a show on antibiotics, which is one of the hardest things I've done. I, I had to cram 2 medical texts to get that thing right. But, but I am prone to such things.
Leo Laporte [01:45:35]:
Paul, of course, has a website, tharott.com.
Paul Thurrott [01:45:37]:
Paul has a website.
Leo Laporte [01:45:39]:
We all wish him well.
Richard Campbell [01:45:41]:
We love his website.
Paul Thurrott [01:45:43]:
It's cute.
Steve Gibson [01:45:44]:
Unfortunately, no one knows how to spell it.
Paul Thurrott [01:45:45]:
Yeah, exactly.
Leo Laporte [01:45:46]:
No, it's a great website. You should become a premium member. You'll get copies of Paul's books if you do, and you also get access to additional content. But it is the website to go to if you want to see what's going on with Microsoft, and you write it all up together. They do Windows Weekly, which is normally every Wednesday, 11 AM Pacific, 2 PM Eastern. You can Tune in, watch us live, or download it from twit.tv/ww. If you're interested in Microsoft, that's the show. So glad you guys could come to Las Vegas and do this show with us.
Leo Laporte [01:46:13]:
Steve, Steve, of course.
Paul Thurrott [01:46:15]:
I only did it to see Steve. I don't really— I know the rest of it, I don't care.
Leo Laporte [01:46:19]:
I know. You know what, let's, let's go to— all go to dinner together.
Richard Campbell [01:46:23]:
I love that.
Steve Gibson [01:46:24]:
All right, I think we have a steakhouse, and the ladies will be all spies.
Paul Thurrott [01:46:28]:
I know, they're all gonna be shiny. Very Very sad.
Richard Campbell [01:46:32]:
Yeah.
Leo Laporte [01:46:32]:
Actually, my wife, poor Elisa, is hurting fans at this point. So she's maybe less relaxed. Steve is at GRC.com. His bread and butter— I talked to somebody the other day who's been a SpinRite owner. So I take a tai chi class, right?
Richard Campbell [01:46:48]:
Yeah.
Leo Laporte [01:46:49]:
And this guy, I've been in this tai chi class with this guy for 2 years. And he said he used to work in security. He used to do firmware. for— oh, we've been talking about it. Was it Fortinet? It's one of the hardware devices.
Steve Gibson [01:47:02]:
Yeah.
Leo Laporte [01:47:03]:
And he would write the firmware. And for the first time in 2 years, he said, well, have fun in Vegas with Steve. And I said, what? How'd you know that? He said, well, I listen to Security Now. He's a fan.
Paul Thurrott [01:47:14]:
Nice.
Leo Laporte [01:47:15]:
And he said, I have SpinRite. I've had SpinRite for 30 years. And I said, isn't it great? Steve gives you free updates. You continue to get free updates. He says, yep, I've got 6.1. If you have a hard drive, you have mass storage of any kind, SSD, and nowadays if you have an SSD, you want to take really good care of it, you should have SpinRite.
Steve Gibson [01:47:34]:
It's gold.
Leo Laporte [01:47:35]:
It's gold. The world's best—
Paul Thurrott [01:47:37]:
Worth its weight in material. Yeah.
Leo Laporte [01:47:39]:
Mass storage.
Richard Campbell [01:47:40]:
Difficult to replace.
Leo Laporte [01:47:41]:
Mass storage enhancement, performance enhancement, repair, maintenance. You got to have it and it'll keep your SSD running for a long time.
Steve Gibson [01:47:50]:
Keep it going fast.
Leo Laporte [01:47:51]:
Yeah.
Paul Thurrott [01:47:51]:
Yep.
Steve Gibson [01:47:51]:
Yep.
Leo Laporte [01:47:52]:
He also does a really nice program, the DNS— we were talking about it earlier, the DNS Benchmark Pro. Yes, it's a Windows program. Well, as all programs should be. Maybe it will be a Swift program soon. We don't know. But you can get that also at grc.com. Now, if you want to send Steve email, the kinds of questions we've been answering today, grc.com/email. You need to whitelist your email address, before you email him or it'll just go in the spam bucket, but he has a way of verifying it.
Leo Laporte [01:48:23]:
Right below that—
Paul Thurrott [01:48:24]:
I need this.
Leo Laporte [01:48:25]:
I know, everybody needs that.
Paul Thurrott [01:48:26]:
I need this.
Leo Laporte [01:48:27]:
I mean, shush, I was telling Steve, I can't find anything in email anymore. Right below that, there are 2 checkboxes. If you want more email, there are 2 checkboxes. One is the weekly show notes. Steve works very hard, usually 20+ pages of information, links, photos, everything. everything we do on the show. He will send that to you every week. And below that, there is a checkbox for a mailing list he never will send you anything from because it only comes out when he's got a new product, which is pretty much never.
Richard Campbell [01:48:55]:
Pretty rare.
Leo Laporte [01:48:56]:
So it's a rare and it's a wonderful welcome gift when the mail arrives from Steve.
Richard Campbell [01:49:02]:
Nice.
Leo Laporte [01:49:03]:
So GRC.com for that. He also has copies of the show, Security Now, this show. He has a 16-kilobit audio version, which no one should listen to. Well, if you have ears, you should listen to it.
Steve Gibson [01:49:13]:
It's the scratchy record version.
Richard Campbell [01:49:14]:
Yeah, but it's It's small, it has a kilobit.
Leo Laporte [01:49:16]:
Yeah, it has a virtue in being very, very small.
Steve Gibson [01:49:19]:
Elaine, who does the transcriptions, used to have a bandwidth-throttled satellite link, right? And so she needed to really budget her bits.
Leo Laporte [01:49:29]:
What was that called? HughesNet? Or what was the name of the—
Richard Campbell [01:49:32]:
a couple of— HughesNet.
Leo Laporte [01:49:33]:
HughesNet.
Richard Campbell [01:49:34]:
I did a bunch of those.
Leo Laporte [01:49:35]:
Yeah, and they had that fair use policy, which meant don't use the don't use policy, and then you could always have bandwidth Anyway, 16 kilobit, but also 64 kilobit, which sounds just fine. And the show notes are all available at grc.com. We keep this show also at our website, twit.tv/SN. We do stream it live. And right now we have, we have about 500 people been watching us through the show live.
Richard Campbell [01:49:57]:
Cool.
Steve Gibson [01:49:57]:
Wow.
Leo Laporte [01:49:58]:
On YouTube, Twitch, x.com, Facebook, LinkedIn, Kick. We see all of you. Thank you. It's great to see you here. Thank you for being here, except for that one scammer. On Facebook.
Richard Campbell [01:50:08]:
Did you see that one?
Leo Laporte [01:50:09]:
Yeah, he said, hey, if you're having trouble losing any— would you lose your information? We can help you. And I thought, no, bye-bye, don't. I hope they kicked him out. Oh man, uh, they're everywhere, folks.
Steve Gibson [01:50:21]:
And I see that, uh, we have a comment: Steve doesn't get any spam and Paul is ready to hit the bar.
Paul Thurrott [01:50:26]:
Both of those things are true.
Leo Laporte [01:50:27]:
Yes, but just follow— if you want the best whiskey, just follow Richard. He knows where Kept somewhere.
Richard Campbell [01:50:33]:
We had a new one last night.
Leo Laporte [01:50:35]:
Something weird in his closet.
Paul Thurrott [01:50:36]:
I think I'm looking for something in a teeny.
Leo Laporte [01:50:39]:
Yeah, I bet you are. Um, uh, now I've lost my thread. Oh yeah, we stream it live. You can watch us live if you're in the club. Of course you can watch live in the club Discord as well. We are very grateful to all of you club members for making this possible.
Steve Gibson [01:50:54]:
Why do I look small compared to the other?
Leo Laporte [01:50:57]:
Are you slumping?
Paul Thurrott [01:50:58]:
I don't know.
Leo Laporte [01:50:58]:
Sit up, old man.
Paul Thurrott [01:50:59]:
Sure, I got a little You get the short chair? Yeah.
Leo Laporte [01:51:02]:
I'm sitting on a stool.
Paul Thurrott [01:51:03]:
Eventually, you'll just be Yoda-sized. Which is actually appropriate.
Leo Laporte [01:51:10]:
He's the Yoda of our show. What else is there to say? But just, I guess, thank you. Thanks to our wonderful team back home at the ranch. John Ashley helped us out. Kevin King, of course, here in the studio. Anthony Nielsen. Benito will be working on the show later.
Paul Thurrott [01:51:25]:
Anthony's done a great job of staying awake I just want to point that out.
Leo Laporte [01:51:28]:
That's amazing. Amazing.
Richard Campbell [01:51:29]:
Absolutely killing it.
Paul Thurrott [01:51:31]:
He did a real nice turnaround.
Leo Laporte [01:51:33]:
Could not hold back when you started talking about physical discs on the PlayStation though, on Windows Weekly. He, he had, he couldn't step in. He had to say something.
Paul Thurrott [01:51:41]:
Got really animated all of a sudden.
Leo Laporte [01:51:43]:
Little passion, little passion going on.
Paul Thurrott [01:51:45]:
Those, the physical media guys are always like that, you know?
Richard Campbell [01:51:47]:
Yeah.
Leo Laporte [01:51:47]:
It's fun for us to get out of that. You know, TWiT is a fully remote operation. None of us are in the same place. The only people who are, are my wife and I, Lisa and I, and that's it.
Paul Thurrott [01:51:56]:
Right.
Leo Laporte [01:51:56]:
And sometimes we wish we were in different locations. So no, we have—
Steve Gibson [01:52:00]:
From each other?
Leo Laporte [01:52:03]:
No, yeah, we want to do the whole show in Cabo San Lucas.
Paul Thurrott [01:52:05]:
Oh, I see.
Leo Laporte [01:52:06]:
That location. No, yeah, but it is a remote group, and so it's nice when we can get together and do some stuff like this. I think it really makes it so much fun. It's really fun to come to trade shows too. As you know, we were— Paul and I were talking about this I love this.
Steve Gibson [01:52:20]:
It's like nothing else like it.
Paul Thurrott [01:52:22]:
Yeah, well, it's, it's going to the show but not going to the show. Yeah, this is honestly kind of awesome.
Steve Gibson [01:52:28]:
At the show. Yeah, be at the show but not actually go to the show.
Leo Laporte [01:52:32]:
Yes, exactly. Ah, there we go. Now we got a shot of, uh, it's—
Paul Thurrott [01:52:36]:
there you go.
Leo Laporte [01:52:36]:
It is— this is just one of several halls. This is the business hall, uh, and of course, uh, tomorrow the hackers come to town. Yeah, because DEF CON is DEF CON, and that will be— I've never done a DEF CON. I've always wanted to. Oh wow, it'll be very interesting.
Richard Campbell [01:52:51]:
Yeah, I'm gonna go home.
Leo Laporte [01:52:53]:
Yeah, yeah, I think we did it all. Get out of here while the getting's good. Thank you, Anthony Nielsen. Appreciate the hard work you did. He, he's got this whole gear, got it all set up.
Richard Campbell [01:53:03]:
That's great, right?
Leo Laporte [01:53:04]:
Um, and yeah, I mean, it's a portable small rig. We have a mobile rig now. We even have an on-the-air battery-powered on-the-air light. We were concerned we would be rushed. That turned out not to be a problem.
Richard Campbell [01:53:16]:
No, we did well.
Leo Laporte [01:53:18]:
All right, so thank you everybody. We appreciate it. We will see you soon. Come back and join us next week, next Tuesday for Security Now. We'll be back at our regular time.
Steve Gibson [01:53:27]:
And the day after for Windows Weekly.
Leo Laporte [01:53:28]:
Windows Weekly on Wednesday.
Paul Thurrott [01:53:31]:
Good to see you, sir. Likewise.
Leo Laporte [01:53:32]:
Take care.
Richard Campbell [01:53:33]:
Good fun.
Leo Laporte [01:53:34]:
Security Now.
