Anthropic warns that AI models could self-replicate. Here's what to know.

Video Transcript

Anthropic had had a warning out right late yesterday about how these AI models could basically uh infiltrate, you know, internet of things or IoT devices like Android TV, uh like they said the fridges, that's like Samsung Smart fridges,

microwaves

use the compute microwaves use that are connected, right? Use the computing power to run themselves and self-replicate. Uh in this case, if this happened and the AI went rogue, uh Ethic says it'd be quote challenging to shut down since there's no centralized point of control to shut off, right? So they could

infiltrate in kind of you know, in aggressive and not uh uh a way that you weren't you didn't authorize it to do so.

It takes over your let's I'm not going to say Apple TV because Apple products seem to be pretty locked down from a, you know, a security point of view. But another type of

Modem, they said they could just get to modem an internet modem.

but that just to be clearer, they don't seem to be saying this is not a hugging face situation. This is not a situation where they say it can get out of our control and this can happen. They seem to be saying is that bad actors

could do this

if they are trying to do this, this this could be one way that even if you tried to shut them off at the data center source, that it could, you know, the so-called kill switch could not necessarily work if there was a hack in progress, it could then go to these home devices.

these like hackers already do that with viruses that can that can, you know, it can go into other types of devices like this already. Uh and and do these things and do sort of like DDoS attacks and stuff like that. But in this case, it would be agents self-replicating themselves, right? potentially which could even be more damaging, right? per this per this warning.

Well, to your point, IoT devices are already huge targets for state security services who are trying to get at somebody. So if now you can push that even further with some kind of agentic system, then that's the fear.

We just saw this in France, right? You brought it up to me.

Yes, we did. So there was a massive hack uh throughout France and I believe the Netherlands. Uh it seems some kind of rogue group hacking into uh uh what are the grocery store systems, refrigeration systems to turn up the heat or turn off the freezing. So employees in in Paris and cities kind of around came in in the morning to kind of restock the shelves, opened up the fridge and got blasted with a wave of hot air that has already resulted in millions of dollars in losses because all that food is spoiled.

That is a really serious threat. This comes maybe what, two, three weeks after we got a report from the Dutch Security Service, which I brought up uh and I don't think I said it on here, but in a meeting with us, this exact point that you can like these hackers can now get into cars and get into vehicles and any kind of smart car you're driving, whether it's a Tesla or anything with a smart system, you have to assume it is always listening and always feeding everything back to a centralized space.