AI’s Hacking Crisis, Anthropic’s Wealth Pledge, Kung Fu Robots

Justin: all of the vulnerabilities that exist today are going to be found, publicized, and patched

Frank: I, I, I don't see it. I don't see it. but I'm glad you're so optimistic

Justin: Hello. Good morning, good afternoon, and if I don't see you later, good evening, which in the current climate may actually happen. Welcome to the AI Argument. The sun is shining, the weather is good, all is well in the world. You're joined by myself, Justin Collery, the techno, some say optimist, I say realist, and Frank, the ever careful, I don't know, techno, technocarian, we'll call you from now on.

Um, no

Frank: I don't even know what that means, but I, I li- but I'll take it, yeah.

Justin: Yeah, yeah. So no Singularity update this week because the world is on fire with, um, hacks and exploits, and the internet is burning down. And I'm sensing, Frank, that you are going to say that you were right all along. But I'm-- We'll, let's, we're gonna do the, let's do the hero's journey on this, right?

We'll have to go down into the depths of depression, and then at the end, I'll tell you why it's actually all a good thing, and it's all good.

PHP vulnerabilities: how worried should we be?

Justin: So let's just describe what happened this week. Do you

Frank: Yeah. So, well, so I mean, the, the, the huge news was, um, was it the day before yesterday? I've lost track of this week, it's been so crazy. Uh, I think it was the day before yesterday, the-- there was a-- there were two critical vulnerabilities discovered, uh, in P- uh, in PHP or, or announced rather, announced in PHP, and there was a huge scrabble to get the vulnerabilities patched.

Um, there had been, uh... So yeah, the, the, the, the, i- for anyone who doesn't know, PHP is an open source code base and about 70% of the internet runs on PHP. So for example, in terms of like, say, small businesses, if your website is, is running WordPress, that runs on PHP, and you would need to, uh, A, get your PHP version updated to a version that could be patched for these vulnerabilities, uh, and your host had to patch that version of PHP for you as well.

So

Justin: Let me just

Frank: was scrabbling to get these patches in place and make sure that their websites were secure because these vulnerabilities allowed someone to do stuff on your website without needing any login details.

Justin: All right. So, and to get, to maybe sort of give a sense to people who may not be close to this, uh, to the seriousness, um, that security professionals and governments were placing on this, Japan declared a state of emergency as a result of this. Uh, the UK government and the European Union suggested that companies who hadn't yet patched their software actually took down their websites, take them offline until this vulnerability was, was, um, was caught.

But what made this week so special was that it wasn't just one thing that happened. There was a whole raft of security holes and, um, you know, issues, you know, any one of which would've been the biggest news of the year in the security world, but there were many of them.

Can AI find flaws before the hackers?

Justin: And these all, all these security holes are being found by AI, right?

So AI is ... Uh, in a, in a, by the way, it should be pointed out that these, all of the vulnerabilities that we've found so far have been found by the good guys. So they've been found by the defenders, uh, using models that have been released by OpenAI and Anthropic, and they're working with the companies to patch the software and fix them.

But the fear, of course, is that at some point, you know, the baddies, the bad actors are, are gonna find them before the good guys find these vulnerabilities

Frank: Yeah, and in terms of the PHP one, it was a very close call because excuse me, I think, um, they, they were basically concerned that either bad actors had found the vulnerabilities just before they were able to actually get the patches out into the world, or concerned that maybe, um, the vulnerabilities had been leaked from the organizations that had discovered them, um, because there were attacks spotted before the vulnerabil- before the announcement was made and before the patches were made.

In fact, I think they had to bring the announcement forward to make people aware because they started seeing some attacks

Justin: Yeah. So, and to that point, right, CloudFare, uh, Cloudflare I should say, is a big, um, sort of web provider, right? It's a company that hosts loads of websites and, and, and serves out these websites. So if you're logging onto a website, you're quite possibly using some Cloudflare infrastructure. And they released some statistics.

They said that 10 hours before it was publicly announced, they saw a few thousands of attempts to exploit this loophole. So that jumped to tens of thousands in the 45 minutes after the announcement, hundreds of thousands within hours, and millions now. And so, so it's just exploded with the number of bots and AIs that people are running locally trying to find websites that have this.

And that's what makes this period in time different, because up until now, a lot of security, um, and the way that these things have been patched have been based on the human ability to do work, right? So if it takes a long time to ... This, by the way, this PHP exploit isn't new. Like it's not something that's been sort of introduced because of some AI slop code that was released.

It's a piece of code that was written by a human and has existed for I think 10 or 12 years. So it's been there latent in the software the whole time. It's just that I think two years ago there was-- Actually, two years ago somebody did say, "Here, look, that thing looks like a, a vulnerability," but they weren't able to put together an exploit that actually could take advantage of the vulnerability to do anything, right, that you wouldn't want an external party to do.

The difference now

Frank: Yeah, exactly, which is where AI comes in

Justin: Yes, the difference now is that we have AI, which is just ultimately persistent. It'll just keep on going. It doesn't get tired. And so these sort of defenses where it's almost security through obscurity, so, you know, the thing is just so hard to do, no human is ever likely to do, those rules don't apply anymore, and that's what makes the current period so dangerous.

How many more vulnerabilities were found?

Justin: Now, just to, I mean, the PHP thing, right, which is, as you say, I mean, an enormous percentage of the internet, uh, runs on PHP, an enormous number of the websites that you connect to, uh, run on PHP. But there's many other attacks that have been going on this week that have been just unusual in their, in their veracity.

So Microsoft reported that there was hundreds of agents attacking hundreds of organizations within their Microsoft Azure, um, uh, infrastructure, and that some of these agents within three hours had gained root access to, um, the, the accounts which they were attacking. Um, CrowdShare, uh, CrowdStrike had documented four operators running successful AI agent campaigns.

One compromised at least 15 Argentinian healthcare organizations. Railway, which is a developer platform that helps p- It's kind of like a low-code, um, this is a big one by the way. It's a low-code type platform which makes you, uh, basically non-techies write software more easily, which is a really big attack surface.

They got successfully hacked, and all of the people's API keys. So an API key, if you think about it, is kind of like your credit card because people can use your API key and spend your money once they have it. So everybody's, or a large proportion or a large number of API keys were leaked from there.

GitLab found a security flaw that existed for 12 years, 12 years that it existed, and they had to patch that. Um, and then there was your favorite one. Your favorite one was the certificate from an Indian company which, uh, led to, um, I think it's, they could hack websites. So there was a, there was a cert that wasn't pr- correctly protected.

Frank: Yeah, so they were s- I think they were meant to be using cryptographic keys, which I think should have been, like, about 100 characters long and, like, 256 characters long, randomly, randomly selected. Uh, so yeah, imagine trying to crack that password. Um, but instead, one of the cryptographic private keys was literally the number four.

Justin: Yes.

Frank: And

Justin: Nobody would guess that. Nobody could guess that one.

Frank: and again, where, where AI comes into this is that the researcher who spotted this, uh, I think he, he spotted it using regular security tools, but then in order to figure out how serious this was, he just gave a really simple prompt to his local AI agent and said, "Can you get me the cryptographic keys there for that cert?"

And they were like, "Yeah, sure. Here you go. It's the number four."

Justin: So, so let me take you to, let me take you... That's just mad. So let me

Frank: So just like, in other words, like you were, you were saying, you know, with the PHP one, yes, it's very serious. It's been there for, for years, and the problem is now AI can find it and exploit it, and it's the same with the eMudhra one. It's the same with a lot of these, that a lot of these might have been from human error.

They've just been hidden away in a pile of code that nobody would ever notice. And but now I could literally sit down and t- ask an AI agent, uh, "Could you find an exploit there and get me into this website?" And it might very well be able to do so

Justin: Yes.

Could AI turn your car into a hacker?

Justin: So let me take you down a rabbit hole of why this appears to be so scary at this particular moment in time, right? Finding a vulnerability in PHP is, is fine, right? That's cool. Um, but it's at least to their web servers, right? And people can connect to them, which means that they're easily accessible and, right, they can be patched and fixed and upgraded and stuff like that.

But let's look a little bit at the history of AI in the last couple of years, right? We went from the point where two or three years ago, AI could write a couple of lines of code, then it could write a whole function, then it could write a program, and now we're at the point where you can leave it running for 12 hours and it'll write an entire system for you reasonably well.

So just, it's the trajectory, right, that I'm interested in. Then you look at, uh, maths problems, so the number of millennium maths prizes or Erdős conjectures or whatever you wanna pick solved in, you know, June was one, by August it was four, September it was 80, and then just the last couple of weeks it was 712, I believe it was.

So again, you can see the trajectory. And so you have to think of the same type of trajectory when it comes to these exploits and hacking and vulnerable sy- um, systems that over the next, you know, six, 12 months, these types of attacks are going to become much more severe. Let me frighten you a little bit on a Friday.

I haven't did a Fri- Friday, Frighten Frank Friday in a while, right? So let me give you an example, right? All cars, all cars basically are computers these days, right? They all have computers in them. And you can imagine that a car is, you know, designed really well to be safe for you, but the computer system itself may have vulnerabilities in it, right?

It's, a car is not designed to be, you know, super, super secure. But let's say I can find a vulnerability, I can hack your car. Your car now becomes a mobile hacking device driving around, and it can connect to all sorts of things. And you could imagine, right, within an electricity grid or a power grid or something like that, there's going to be lots of devices which sit on poles or under the ground or whatever.

They're in pumping stations, and they all have computers in them as well. And now that I have, uh, access to your car, I can use your car as a jumping off point maybe to hack an electricity system or a power plant or a water system or whatever it happens to be. And those things are much harder to get access to than a PHP web server, and they have a much greater impact on people, normies like you and me, if those things were to get hacked.

The other danger I see, uh, with this-- Yeah, go on

Frank: just to ex- extend that another little bit to, uh, scare people like me even further, Anth- Anthropic, like, just released a report sta- saying that smart devices, LL- AIs could-- would-- might be able to replicate themselves onto smart devices, and large models might even be able to replicate themselves onto multiple smart devices.

I can't remember the word they used for it, but basically kind of like fragment themselves up across multiple smart devices. So,

Justin: It's a lot of toasters. It's a lot of toasters if you want to run a large language model. But anyway, maybe

Frank: Uh, but like that is insane. Like, so not only, not only could they be hacked, not only could they be used in hacks, but they could actually have a level of intelligence running on them to perform those hacks. When you couple that with another story I read, which was that the US Army were deliberately looking into how they could basically send swarms of agents off to their adversaries to take out their computer systems, and then you think about how even benign tasks that OpenAI were giving their models ended up in a hack on Hugging Face.

Y- now we're looking at a situation where we're actually actively thinking about putting these swarms of agents out into the internet that have tasks that involve hacking, so that is-- that's what they're designed for, but they may not, like, follow your instructions exactly, and they could easily, for one reason or another, turn against you as well, get onto your smart devices, and suddenly we have, you know, we have a world that is just proliferated with agents that are hacking everything

Could your AI agent hack a bank for you?

Justin: So, and to make, to make your Friday even better, Frank, you don't even need to be a bad actor for this to happen. So banks in the US are now very concerned because they're getting loads and loads of traffic from agents, right? And they, they didn't classify whether the agents are bad actors or whether they're just Justin's agent, which goes in and checks his bank statements every month and, you know, downloads it and compiles a report and says, "Here's the state of your finances."

But the problem, I guess, or the potential problem here is, let's say I turn around to my agent and I said, "You know what? I'd really like to buy a new car, preferably one that can't be hacked and take over the water system. And, um, I'm gonna need a loan, so can you go and get me a loan, please? Can you just go and sort it out?

Just get me a loan for 30 grand there, right? And let's go buy a new car." And my agent takes my instructions a little bit too literally, so it connects to the bank, it applies for the loan, the loan gets denied, and it goes, "Yeah, but Justin really wants a new car." And so it starts to, not in a malicious way, hack the bank,

Frank: Well, in a malicious way, just, just in a malicious way, but just not with a malicious intent from on your part. Yeah.

Justin: So we've gone down this dark hole, right? And I do agree, right? This is kind of... 'Cause you said,

Frank: so before, before you come on to the next segment of your thesis, can I just ask you, like at this point of the thesis and at this point in time, how are you feeling about your whack-a-mole approach to AI problems in terms of just whacking them and patching them as they arise?

Can AI patch its way out of this mess?

Justin: So I feel good, and in fact, I think this is a necessary evil. Why? The alternative will be worse. Okay, let me put it first like this, right, before I get into the, the, the, the good side of the thesis, right? This week, and possibly for the next six months, if you were a bad actor, and in particularly if you were a state bad actor, depends which state it is, I guess, right?

But if you wanted to cause mischief, now would be a great time to do it because all these agents are running around and it's-- everything's a little bit up in the air, right? So if I was looking to cause mischief, right, and I was a state actor, now is a great time to do it, right? Total deniability. So if these models weren't released to the public, if we didn't have access to these models, then a small number of people would have the power to cause that mischief.

Whereas what's happening now is more similar to open source software, where everybody is learning about these vulnerabilities and everybody's fixing-- we're whacking the moles. It's the only way that I think you can tackle this. So why is this a good thing? We've gone through, we've gone through this before, right?

We had this thing called Y2K many, many years ago, where everybody thought that all the computer systems were gonna die on the 1st of January in the year 2000. And there was big efforts and a lot of work was put in, and we secured all the systems, and Y2K came and passed, and there wasn't even a blip, right?

Nothing bad happened. And we're about to go through exactly the same process again. The bugs that we are finding at the moment aren't written by AI slop, right? They were written by humans. Um, and they exist in the software. They're latent. They're there already. And we're about to go through a period, right, two good outcomes are gonna come from this, right?

The first is that all of the vulnerabilities that exist today are going to be found, publicized, and patched, and that process is gonna be difficult, but all good processes are difficult, right? That will go on, and it'll take twelve, 18, maybe even 24 months. The second good outcome of this is that as we write more and more software with AI, the AI becomes aware of the vulnerabilities that we had to patch in the past, and it means that all future software that's written should never have the vulnerabilities that we can see existing today.

So you end up patching the existing software and fixing the future software because of this process, and there's no other way that you can do it.

Is AI hacking really another Y2K?

Frank: So two, two pushbacks here. Number one, your Y2K analogy. Y2K, we were all afraid that, that the world was gonna end as the, as the bell tolled on the, on New Year's Eve. Uh, and that was, in essence, as a result of one issue

Justin: Yes, in many systems. It's totally analogous

Frank: But one issue. And now we are already just this week seeing that curve that you, that exponential curve that you talked about, where w- it's not one issue.

Like the PHP issue that, that we were talking about today is one of, as you pointed out, a multitude of issues just this week. So I'm seeing, I'm actually, I'm seeing a lot of people, I saw Barry Scanlon on LinkedIn saying it today, uh, I've seen various people liken this to, they're, they're saying that they feel like this was, this is 2020 and people are starting to get sick, but nobody's really taking it seriously yet.

So

Justin: But we already have-- But we also already ha- we already also have the vaccine, the LLMs to patch the code. So it is a bit like 2020.

Frank: well, you believe we have the vaccine, but so that's my second pushback. My second pushback is you're saying everything will get patched, but if we just take the PHP vulnerabilities, right? Yes, the, you know, for example, I, I immediately contacted the host that I use for any WordPress websites, um, that I have, and they said, "Yeah, we're on it.

It's all good. Um, PHP has been patched, and your sites are on an updated version of PHP. You're all good." So that's fine. But what about all-- And, and sorry, similarly, you know, uh, we can assume that the, you know, the, the big tech companies are all on this. They're all patching their, their systems, et cetera.

Brilliant. But what about all of the abandoned websites that are out there? What about all the things that people have walked away from? What about the projects people have forgotten about? Like, I don't think that we can just say, "Oh yeah, everything, you know, we, we found the vulnerabilities and we found the patch and now so everything's gonna be fine."

And similarly, tomorrow's vulnerability, it might take us longer to find a patch or there may be no patch or, you know-

Justin: No. Okay, so, ah, you bring me nicely onto my next point of why this is gonna be okay. If you think about this attack, right? On the one hand, you've got LLMs, which are non-deterministic and very, very persistent, right? And they will attack, and they'll probe every side of your system. On the other side, the defending side, you have deterministic software.

Software, at the end of the day, is deterministic. It's bits. It's ones and zeros. At the moment, the, the, the advantage is with the attackers because you've got unpatched deterministic software and a, and a thing which can probe it. But at the end of the day, this is deterministic, right? If you put the same input, you will always get the same output.

So over a period of time, this becomes hardened, and this becomes unhackable. And it doesn't matter how good your LLM is, so long as this piece of software doesn't have any vulnerabilities. So long as this is deterministic and this part is non-deterministic, in the long run, the advantage will be with the defender

Frank: I, I, I don't see it. I don't see it. but I'm glad you're so optimistic.

What can we do to protect ourselves?

Justin: Sorry, I should just say, by the way, 'cause this is, for a lot of people this might seem, you know, why does this matter? You know, whatever. Quest Diagnostics, they're a very big US, uh, diagnostics company. I believe that they've done over 200 million tests, so that's maybe 50% of all Americans. They were attacked, they were breached

so it does have real world impacts. So, if you have a, a computer system and your computer system says, you know, you've got an update to your software, update your software. Don't be using the same password in lots of different websites. Change your passwords regularly. If you get a notification from a service that you use,

Frank: don't use a cryptographic key of the number four

Justin: Oh dear, yes, yes, do. So we need to learn-- I, I will say, right, just to make you laugh, um, there was the, a number of researchers, so Joshua Akkam, who used to work at OpenAI, he tweeted about this event during the week. He said, "I do think we ought to seriously consider printing paper records for everything important.

We need tamper-proof logs of everything ASAP in case a lot of critical systems go offline for a time." Uh-oh. And Jason Clinton, who, uh, is one of the security researchers in Anthropic, was tweeting about it, and his tweet was, "Yes, it was always going to be this way. We tried to warn the world." Which isn't, I will admit, not the most comforting message to come, but, but I do think it's gonna be okay

Frank: that reminds me that the, the, the printing paper thing immediately I was like, yeah, or like l- you know, let's all just, let's all just start goat farms.

Do we need a human-only internet?

Frank: Um, and that actually reminded me, I mentioned earlier that I was seeing various people saying, likening this to the, the vibe in 2020 when, when COVID was just kind of, uh, just emerging.

Um, and Wes Roth had a video, and he was again talking about all these hacks, and he was saying, "Yeah, maybe, maybe this is the start of like the exact opposite of COVID," where in COVID there was the virus in the real world, and we all ended up going online. We all, we were all on Zoom and, and online shopping went crazy, and he was like, "This could be the opposite.

This could be the internet is, is hacked and vulnerable and, uh, collapsing, and we're all forced back into the real world."

Justin: Yeah, it takes me back. We did think about an idea for different reasons a while ago. Uh, we were worried about the dead internet theory and about the internet just being so full of AI slop that, you know, it became worthless. And here's another reason, all right, and the, the solution to that at the time, one of the solutions that we thought about was that we-- you separated the internet, so you had an internet which included AI, and you had another internet which was humans only.

If only there was an orb or something that could scan my eyeball to prove that I was human, but there we go, that one, that dream died last week. Um, and this is yet another reason why you might want to have-- I don't have a good technical solution as to how you'd make that happen since there's no orbs, but, um, you know, maybe a human-only internet and a human and agentic internet, maybe they'll bifurcate and you'll have both of them at some point in the future.

Is Musk putting money where his mouth is?

Justin: Anyway, on the upside, we're all gonna get loads of money

Frank: Yeah, we're, we're all gonna be rich. We're, uh, yeah, universal thriving income is, is on the horizon

Justin: Brilliant. Two big announcements this week, one from Anthropic and one from Elon Musk. None of them benefit us because we're Europeans.

Frank: Well, we don't know. We don't know that.

Justin: Really?

Frank: Well, we don't, we don't know about... Yeah, tell us about Elon Musk

Justin: Elon Musk is gonna give one share from XAI/SpaceX/Tesla, I think they're gonna merge soon, to every child in the United States, but only the United States.

So, uh, good for them, and hope-

Frank: uh, and I admit I'm surprised and, you know, I, I think it was like right back when we started this show, Elon Musk was talking about how AI would necessitate the need for universal thriving income, that AI would basically take over the commercial layer of, of society and the wealth would be redistributed to people and we'd all be, we'd all live in this wonderful utopian future.

And I was like, "Okay, great. How are you making it happen, Elon?" Like, put your money where your mouth is. And, you know, at least this is a step in the right direction.

Justin: Or at least he said he will. Yes, uh, so that's great. And, you know, maybe, maybe this is the way of the future.

Anthropic's fund: who gets the money?

Justin: Anthropic also made moves and, um, so d- what have Anthropic done this week? Have they done

Frank: Yeah. So they have announced, uh, a public benefit fund, and we've, you know, we talked about rumors about this before, and OpenAI were talking to the government about something vaguely similar. What they've basically said is that they're going to put, like, 5% of their, uh, company equity into this public benefit fund, and that 5% could be increased as the company becomes more valuable.

So it could, you know, e- even at 5%, I think it, it could potentially be enormously va- an enormously valuable fund, and if it grows, it could be, it could be huge. The-- and they've, they've talked about, like, you know, what might those funds then be used for. They might explore large donation, d- large donations to, um, charities, nonprofits.

They also specifically said, and this is where the, the possibility of something like universal basic income, um, kind of makes an appearance, they said they, they might consider direct distributions to specific populations, for example, people whose jobs are impacted by AI. Um, but, you know, again, a little bit like you said there with Elon, that yes, he has said he will, but what actually happens, this is very similar because they've basically said, "We are going to do this."

We don't really have any details of how it's gonna work. It will be an LLC. They will be completely in control of it. Beyond that, we really don't know how it's gonna work, who manages the fund, how much of the fund is actually distributed in terms of cold, hard cash, um, and who gets it or anything like that.

There's no details. They've said they'll have details coming up. So,

Justin: All right

Frank: the more cynical readers of this have been like

Justin: AKA Frank

Frank: and your, your, your old friend Gary Marcus also,

Justin: Oh yeah.

Frank: 'cause I, I, I was kind of-- I was looking into this. I was like, "Could this be a more cynical move here?" And I typed in my theory that maybe they were just trying to get, um, goodwill, public goodwill, uh, ahead of their IPO. Um, and when I, when I put my theory in, it was like, "Well, that's what Gary Marcus thinks."

Justin: There you go. There-- Then you know it's wrong, I would say. Um, here is my take on, on that. What I think about that is that's great, right? And what they're doing is they're showing-- it's mood music, and they're showing a direction of travel. But Anthropic, uh, I think that it's hard, right? And again, it's one of these things where it's like, well, that's great if Anthropic do it, uh, but all the other companies have to do it too if it's to benefit all of humanity, and it has to be spread across all of humanity, not just to one particular country.

The best way to do that is two things to happen. First is that countries like, uh, some countries in the European Union, I would love for it to happen in Ireland, should build loads of electricity power stations and loads of data centers, 'cause they will be the center of gravity of power in the future.

Then you make those available to Anthropic and OpenAI and xAI and Google and anybody else that wants to serve out their models, and then you tax the electricity that's been used by the, uh, the, the data centers, and that's what you use then to spread your universal thriving income. That way, it doesn't matter which company uses the compute, it doesn't matter which company has which idea as to which is the best way to spread out the money.

The money is controlled by the governments who are elected by you and me, and we decide what best way to spread the money. So

Frank: Yeah. I

Justin: I

think it's the right move, but-

Frank: Yeah, I think you're right. I think, I think it's, it's directionally, it's really good. I do think that Dario Amodei genuinely believes, you know, what he's doing. Um, they've already pledged 80% of their wealth, um, the, the founders of Anthropic have already pledged 80% of their wealth to, um, that, that they will donate it.

Um, and so I do believe he, he believes this, and I think you're probably right. He's like, he's, he's trying to lead the way here. He's trying to show the way. But my-- Yeah, I think exactly what you've said, my old refrain, it should not be up to the companies themselves.

Justin: Wow, we're in agreement. Incredible.

Frank: Amazing

Why does Unitree's robot hate cars?

Justin: one of my favorite, uh, robots ever to be created, and I-- we haven't seen enough of it recently, is the Thermonator, which was one of those robot dog type things with a flamethrower on top of it, which was just absolutely fantastic. And with all of this hacking going on and, you know, the danger of civil unrest, France is on fire, Unitree, our favorite Chinese, I think they're Chinese, uh, robot company, has released a robot to put yours and my mind at rest should the worst happen

Frank: Indeed, yes. Uh, th- this is a, I guess you, I don't know, it's a, it's like a kung fu robot that they're saying will basically act like your bodyguard and protect you. And from what I can see, we'll, we'll, let's play the, let's play the

Justin: Kung Fu Legend. I mean, there we go. Okay

Frank: there, yeah, it literally says kung fu legend, yes. Um, and it looks like

Justin: All right, so wow. Did you see that? It just did a twirl. Oh, it doesn't like the car.

Frank: He is attacking that car with gusto. Another car he's attacking. Right. So are they worried about people being attacked by cars?

Justin: I think this is how you get to the baddie

Frank: Why is he in a car scrapyard? This robot hates cars.

Justin: These are all the cars that he destroyed previously. It's like a graveyard of cars.

Frank: All the... Oh, look at this. There's a guy coming in with a knife, and the guy with the knife pauses dramatically to give the robot a chance to get in and protect the human. Okay, somebody has been run over by a car. Robot lifts the car so someone can pull out the person. Oh, someone's about to be run over by a car. Man, the cars are the villains here. What is going

Justin: I tell you? I tell you. So

Frank: wait, why does it end with a, a robot waking up from a nightmare? What is going on? That I-- what was that?

What was that?

Justin: So, and can it make the tea? That's my big question. It's great having your bodyguard that can defend you against all sort of evils, but, you know, what if I'm just chilling, looking out at the sea and I'd like a cup of tea? Can it do that? But look it, I feel better. I would buy one. I don't know

Frank: you, are you, are you afraid of cars?

Justin: I'm afraid of cars. Yeah, I don't know what it has against cars. But, um, look it, who knows? We-- Look it, I mean, according to Elon Musk, there's gonna be billions of these robots roaming the earth within two or three years, so you're gonna need some protection

Frank: Maybe, maybe they were all Teslas. Maybe they were all Teslas that had been taken over by rogue AI agents and were attacking the humans, and so you need a unitary kung fu robot to protect you from these rogue agents that have infected cars. That's clearly what was going on. That's clearly what was going

Justin: it's ... See? We're on the same page. Brilliant. Right, if there's an internet next week, I look forward to talking to you then. If not, it's been great knowing you.

Frank: Excellent chat too. Maybe