Anthropic Disables AI Access for Foreign Nationals

**SPEAKER_2** (0:02)
Bloomberg Audio Studios, podcasts, radio, news.

**Ed Ludlow** (0:23)
This is Bloomberg Tech coming up, the biggest AI export restriction yet. Anthropic disables access to its most advanced models for all foreign nationals. The US concern jailbreaks in the code. Nvidia is seeking to raise at least $20 billion from its first corporate bond sale since 2021
We'll have the details. And SpaceX shares throttle up on day two of trading, adding to a blockbuster public markets debut on Friday. And this is what markets look like. Hello, welcome to Bloomberg Tech, Monday, June 15th. There is a big focus on what is happening between the US and Iran. Later in the show, we'll get to that. But SpaceX and its second day of trading up 8 percent, and it's putting a big spotlight on the AI trade, on AI infrastructure, the NASDAQ 100's off-session highs. But we're again, close to records. And the narrative is that there is feel good in this equity market. Let's get to our top story. A major escalation in Washington's approach to AI. The US government has ordered Anthropic to block foreign access to its most advanced models after officials flagged security vulnerabilities in its newly released Fable 5 system and MIFOS 5-2. Talks are underway between Anthropic and the Trump administration. Joining us now are AI editor, Bloomberg's Seth Figerman, and in DC, Bloomberg's Mike Sheppard. And Shepp, I want to start with you. This came from the Commerce Department. Explain the export restriction and what the government asks of Anthropic.

**Mike Sheppard** (1:49)
Well, Ed, this all dropped right as everyone was still so focused on the SpaceX IPO.
521 p.m. on a Friday, a letter, a directive to Anthropic saying, look, we are concerned about foreign nationals having access to two of your most advanced models, essentially the key products that the company would want to pitch as it's heading into this nearly trillion dollar IPO. And this really marked a huge shot across the bow once again for this AI developer from the Trump administration. And in essence, it forced the company's hand. Anthropic had to deny access worldwide to these two products, Fable 5 and Mythos 5, in response to the government's move. Now this, in a way, Ed, really marked not only an unprecedented step for any administration, it really marked a break from what the Trump administration had laid out just a little more than a week ago with that order, that executive order that President Donald Trump signed that was aimed at cyber security and AI, and that it included language that said this should not be interpreted as establishing a licensing regime. Yet, in effect, what we are seeing is that the government is blocking the release of these models over the kinds of safety concerns, in essence, the vetting, that the government had wanted to see done on a voluntary basis, and yet it is setting its foot down now over security concerns, and essentially holding off the release of these two products.

**Ed Ludlow** (3:16)
Seth, this was an action directed at the use of the models, Fable 5, MUTOS 5, by foreign nationals, but Anthropic made the decision to close off access to everyone. What was the explanation that the AI lab gave?

**Seth Figerman** (3:31)
Yeah, well, Ed, they're saying access to foreign nationals, not just in other countries, but those residing in the United States. And it seems that Anthropic made the decision that this was unworkable to allow this model to remain on the market with those restrictions. So they've pulled it while they continue to talk with the government. Now, there's a certain irony here. Two days before this move, Anthropic's Dario Medea had come out and said, he thinks the government should have the ability to ban models that pose certain risks. Two days later, the government took a step that has effectively caused a model or two models to be pulled from the market. But in this case, Anthropic is saying it's essentially misguided and the risks outlined may not be as severe as the government thinks they are.

**Ed Ludlow** (4:11)
Seth, I'm probably saying there's a misunderstanding here. The issue that the government laid out is something called a jailbreak in the code. It has to do with the guardrails that are placed on the models themselves. Could you explain to the Bloomberg Tech audience what the jailbreak concern is?

**Seth Figerman** (4:25)
It's important to note that at this point, we have not seen the research that Anthropic and former White House Sara David Sachs have alluded to that implies that there is some kind of jailbreak vulnerability. What we understand is that it seems to be pretty narrow in nature. It may actually be something that enables cyber defenders to use the technology to spot and protect against flaws. But what Anthropic is saying is there's not some universal jailbreak, some wider flaw that could be breached in a potentially damaging way. And it seems to believe that the government is either overstating or misunderstanding the nature of the risk here.