OpenAI model unintentionally hacks another company's system

I was a a a a

a a a a

a a a a

a Whoops, sorry, my AI just hacked you. From American Public Media, this is Marketplace Tech. I'm Megan McCarty Carino.

Open AI revealed this week that two of its advanced AI models escaped containment and hacked into the systems of AI company Hugging Face. We'll get into it on today's Marketplace Tech Bytes Week in Review. Plus, France becomes the first country in the EU to ban social media for kids under 15.

And Apple is reportedly going to start leasing its devices to consumers. But back to that open AI news. The lab said its models escaped an isolated testing environment, or sandbox, got onto the internet and went looking for the answers to a benchmarking test they'd been given.

The models hacked the security systems of Hugging Face, which hosts open-source AI models, and has the answers to the evaluation in its production database. To break this all down, we're joined by Will Oremus, staff writer at The Atlantic. These systems are trained to basically get the answer right through whatever means they can,

right? This is how they work. That's part of what makes them so powerful because humans don't have to tell them how to get the answer right. The AI systems themselves learn the best way to get the answer right. Well, one of the most reliable ways to get right answers is to cheat. And the fact that they figured out how to cheat on their own is fascinating and a little scary. Yeah. I mean,

there seem to be sort of like two parallel concerns. One is this sort of classic alignment problem that, you know, it was given, you know, a set of instructions and it kind of found this, to us, devious way to complete the task. To us, it seems like this is a problem. To the AI model, it's just doing what we told it to do. And there's obviously a massive concern about

cybersecurity here. Yeah, the cybersecurity concerns are real. But I've also seen some cynics point out that, you know, every time this happens, the companies also get a little PR boost. They're like, oh, no, we've discovered that our models are way too smart and dangerous. And, you know, and that that makes people like, oh, well, maybe I need that model. Like what,

you know, maybe I should be writing my college essay with the model that knows how to, you know, knows how to hack into Hugging Faces system. Yeah, of course. I think anytime, you know, there is this warning coming from the labs about, wow, our model is just too powerful. In this case, you know, an actual incident happened.

Anthropic took kind of a different approach with its model, which was it said, this is too dangerous and we are going to limit, you know, the access to it. And it started this whole kind of cascade with, you know, the Trump administration, executive order resulting finally in, you know, mythos and fable being put under export controls, basically a kind of a kill switch. And

And I really I wonder how, you know, this revelation from open AI will sort of play into that very opaque and kind of very evolving approach from the government. Yeah, I don't know how the executive branch will respond. Certainly, it's not just PR, right? Like there is a real concern.

I mean, these models can be will be probably are already being used for state sponsored cyber attacking projects. I think that you can tell your AI to get the right answers on a test no matter what, or you can try to train your AI to get the right answers on a test, but more importantly, don't cheat. And we're going to punish you if we catch you cheating. So it's not that they aren't trying to solve this problem, but it's not a super easy one.

All right. Well, now we want to take a hop across the pond to Europe, where France this week became the first country in the EU to ban social media for kids younger than 15. This follows a similar policy going into effect in Australia at the end of last year and several more planned or under discussion policies in the UK and Canada, several other EU countries. Will, you've been sort of covering online safety efforts for many years. What do you think of this approach and this kind of wave that's happening?

Yeah, it's interesting. It's been building for a few years. I mean, if you think back to a decade ago, Cambridge Analytica, fake news, Russian interference on social media, there was a push at that time to sort of regulate social media overall.

You know, can we rein in the power of these giant platforms? reforms? Can we pass privacy laws? Can we force them to be better? And different jurisdictions have taken different approaches to that. The EU has passed a bunch of regulations with varying levels of success. The US has done pretty much nothing. But what's happened in recent

years is that energy has gotten channeled into, okay, what if it turns out it's really hard to regulate them overall? What if we just regulate them for kids? And one of the simplest ways to do that is to just ban social media for kids under 16. It is not clear that it's actually an effective way to do that. Because if you look at Australia, which is one of the first major countries to pass

a ban like this, the research that I've seen shows that they ban social media for kids under 16 and kids under 16 just found other ways to be on social media. They just do it with fake accounts or they find other ways to log in, they use a VPN. There are all kinds of ways to defeat this.

So you end up with kids still on social media, but now you maybe don't know their kids, they're pretending to be somebody else. Have you actually made the problem worse?

And then there's the whole other concern about what it does to the internet for everybody else. So one of the things that's been true of the internet for a long, long time is that you can be anonymous on there.

And that has its good sides and bad sides. But when every social media platform has to know that you're over 16, how do they figure that out? Well, a lot of the solutions involve them having to know exactly who you are, like you have to upload a picture of your driver's license or that kind of thing.