00:00Anthropic says the U.S. government has removed foreign access restrictions on its Fable 5 AI model.
00:06Now for more, let's bring in Bloomberg's Mimmin Lo.
00:08Tell us more about the backdrop and what led to this decision.
00:11So this came about more than a couple of weeks after the U.S. government had told Anthropic
00:15that it needs to seek U.S. permission to have foreign actors access two of its most powerful models.
00:22That's Mythos 5 and Fable 5.
00:24And then a few days ago, we reported that Anthropic had made some changes to address some of the national
00:29security concerns.
00:30And then the U.S. government then lifted some selected restrictions on Mythos 5 to vetted companies.
00:36So things like critical infrastructure providers, cyber defenders.
00:39And we're now learning that for Fable 5, the government has lifted all foreign access restrictions.
00:44And Fable 5 is the less powerful version of Mythos 5.
00:48It's meant to be public-facing and for more widespread use.
00:51So this is a major win for Anthropic and probably going to lift its valuation ahead of that IPO as
00:56well.
00:57Yeah, could be good news for them.
00:58What about the implications?
01:00What could this have, what would it mean for some of Anthropic's Chinese rivals?
01:04Yeah, I think there are twofold.
01:06One is that, first of all, the race is back on with the Chinese rivals.
01:10Because if the U.S. continues to place these restrictions on Anthropic, then it means that any deployment of the
01:16frontier models, whether it's by Anthropic or even OpenAI and all the other U.S. models, that deployment is going
01:22to be very limited if the restrictions are in place, right?
01:25So that hands a lot of advantage to the U.S., to the Chinese players whose prices are already much
01:31lower and that the models are often open source.
01:34So that already increased their market adoption rate.
01:37The other thing is the national security front, right?
01:39Because one of the reasons why the U.S. is clamping down is because they are worried that foreign actors
01:43can circumvent some of these guardrails.
01:45And the question is whether or not that is secure.
01:49Because the U.S. hyperscalers have accused the Chinese ones of a process called distilling, distillation, where you generate a
01:55large amount of data to train your own AI models in a cheaper way.
01:59But this is something that Chinese Premier Li Qiang last week at the World Economic Conference had already hit back
02:04again, saying that you can't copy your way to an innovation edge.
02:08So again, it really raises the question about whether or not the new model by Anthropic could be distilled as
02:14well by these Chinese players.
Comments