The US government just pulled a pretty dramatic move against one of the most prominent AI companies in the country. According to reporting by Reuters, Anthropic was ordered on Friday to “abruptly disable” its most advanced AI models for foreign nationals, citing national security concerns. The models in question are Fable 5 and Mythos 5, the very same ones the company had rolled out earlier this week with much fanfare.
It’s a strange situation when you think about the timing. Anthropic had just launched what it called a new “Mythos-class” tier of capability, positioning these models as a significant leap forward. And now, mere days later, the company is scrambling to cut off access to hundreds of millions of users.
The Government’s Rationale
So what’s the government actually concerned about? According to Anthropic’s statement, the Commerce Department issued an export control directive without providing specific details about the national security threat. The company’s understanding is that officials believe there’s a potential method of bypassing, or “jailbreaking,” a safeguard that would prevent Fable 5 from being used in identifying software vulnerabilities.
Think about that for a second. The government is essentially saying that this model could be coaxed into helping hackers find weaknesses in software systems, and that foreign nationals having access to this capability poses enough of a risk to warrant an emergency shutdown. It’s a striking escalation in how the US approaches AI safety, moving beyondchips and tools to directly restricting access to the models themselves.
The government provided what Anthropic described as only “verbal evidence of a potential narrow, non-universal jailbreak.” That’s a pretty thin foundation for an order that affects basically every user outside the United States.
Anthropic’s Pushback
Unsurprisingly, Anthropic isn’t taking this lying down. The company publicly disputed the basis for the directive, stating that “the finding of a narrow potential jailbreak should not be cause for recalling a commercial model deployed to hundreds of millions of people.” They’re calling it a misunderstanding and say they’re working to restore access as quickly as possible.
There’s also some context here that makes this situation even more complicated. Earlier this year, Anthropic’s relationship with the government soured after it refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems. In response, the government put Anthropic on a supply chain blacklist, set to take effect later in the year. This latest directive feels like another chapter in an increasingly acrimonious standoff.
Former White House official Dean Ball, who contributed to the administration’s AI Action Plan, suggested on social media that the order effectively means all non-Americans would be restricted, including those based in the US. “You should expect to have to prove your citizenship to use Anthropic models,” he posted.
The Bigger Picture
What’s really going on here is bigger than just one company and two models. For years, US export controls have focused on the physical hardware that powers AI, the chips and manufacturing equipment. This order signals a new frontier in that approach, targeting the AI models themselves rather than just the infrastructure.
And here’s where it gets really interesting. Anthropic pointed out that if this standard were applied across the industry, it would basically halt all new model deployments for every frontier AI provider. That’s not hyperbole either. If a narrow potential jailbreak is enough to shut everything down for half the planet, the entire pace of AI development could grind to a crawl under regulatory uncertainty.
The thing is, this isn’t coming from nowhere. Experts have genuinely warned that models like these, in the wrong hands, could accelerate sophisticated cyberattacks, particularly in critical sectors like banking that rely on complex, interconnected systems. Those concerns are real. But the question is whether a blunt instrument like a blanket export ban is the right way to address them, especially when the evidence provided seems so thin.
This whole situation leaves us with a uncomfortable truth: the tension between open AI development and national security concerns isn’t going away. If anything, it’s going to get messier as these models become more powerful and more widely used. The challenge for regulators will be finding ways to address real risks without essentially locking down an entire technology sector. And for companies like Anthropic, the pressure to play nice with government while also maintaining their ethical stances is only going to increase.
