Latest / Black Box Theory Podcast / The Government vs. the AI Labs: Who's Really in Control?
Transcript
- Malcolm: So the government just forced an AI company, Anthropic in this case, to pull its own product and the AI company is now fighting back. ⁓ So three days, ⁓ that is how long Anthropic's most powerful public model to date. ⁓ Claude Fable V. You may you may or may not have heard of it. They did a ton of announcements and rollouts for it. ⁓ if you aren't like super like into Claude and use it super often, you may not have seen it. ⁓ if you use it casually, you may not be familiar. But yeah, Claude Fable V was available before the US government ordered it offline. ⁓ on June 13th, Anthropic received a directive at 5 21 p.m. Eastern ⁓ Standard Time from the federal government invoking export control authorities, ordering the company to immediately cut off all foreign nationals from accessing Fable V and it's Restricted sibling, Mythos V, which is like the most groundbreaking cybersecurity model. ⁓ sorry to interrupt. Is isn't that the one that like released itself early? Yeah, it'll some it's it'd be like breaking out the sandbox. It'd be breaking out the legit like that that's the one that if anything that's the one that need keep on the wraps. If Able Five I I didn't get a chance to use it, bro. ⁓ but yeah, like that includes foreign national anthropic employees inside the United States. ⁓ So since Anthropic can't reliably verify user nationality at scale in real time, the only way to comply was to pull both models globally, everywhere. So now everyone no no one can use it, right? So ⁓ the government stated their reasoning behind it was a jailbreak, specifically a claim that someone could jailbreak Fable Fable V by asking it to read a code-based and identifier software vulnerabilities. Anthropic reviewed the technique ⁓ and pushed back hard. They said that the vulnerabilities identify were minor, already discoverable using other ⁓ publicly available models. ⁓ and if the jailbreak produced no unique uplift, meaning that nothing that Fable Five exposed isn't something that, you know, another model couldn't show or that you couldn't already get elsewhere. ⁓ and so like I say a lot of this evidence from the government is is all verbal. ⁓ there have been no written technical details, no disclosure of a harm harmful outcome. ⁓ but anthropic comply anyway, like the awesome company they are, and the public, you know, of course went with this ⁓ objection, right? So ⁓ yeah, I don't know, man. This is It's interesting. I said this isn't the first time that Anthropic has gotten into it with the federal government. I'm sure another you guys remember like that kind of whole situation with the ⁓ contract with the Pentagon and stuff and kinda like open AI coming in and swooping in after this. But if they say like, you know, the Pentagon's kind of priorities don't really align with our values and and things like that. And that's I think where a lot of people became you know, fans of Anthropic and and kinda like the company and the vision ⁓ that they kind of stand for. So I don't know, man. I think ⁓ hmm I mean, I think we get into that point where it's like, I mean, if the government is starting to get scared that these models are too good, ⁓ It's only a matter of time, bro, before they just start restricting access to everything. It's just like I mean, it it just seems like a very like haphazard like like like a very like just kinda like I thought there's not really much substance behind the I guess accusations or the the reasoning behind having the models pulled. ⁓ saying like, ⁓ it can identify vulnerabilities. I mean I mean they're gonna keep getting better, bro. Like I I don't think that this is like news to anyone. Like I mean if anyone thought it was a joke that these models are gonna continuously improve and that ⁓ like it was all for for for show, I mean you're seeing it now. So I don't know, man. This is ⁓ Miss real time I I I think when it when it comes to the AI story, ⁓ stuff I think the more that the market figures itself out because now we're nearing like token based billing and ⁓ my god yeah that's that's been a thing for a lot of people so expensive ⁓ but as we lay a groundwork foundation for everything I think That structure will allow the government to then build their own structure. Because right now up to you know just a few months ago, AI race has been just a cowboy wild west shootout. Yeah. Like everybody's just throwing everything, moving faster than the government can comprehend. And I mean, all of the brightest minds are going to these companies. The government's not paying people five hundred K plus million dollar in stock and you know, all this other stuff. Like bro the the people are going there. So the government has to catch up. And I think that as the a the industry Settles into a norm of this is how this works, this is how we build, this is how we release, blah, blah, blah, then the government cannot regulate it.