Nine days ago the US government did something unprecedented: it ordered Anthropic to pull its most powerful AI models offline and bar access for every foreign national on Earth.
As of Sunday morning, Claude Fable 5 and Mythos 5 are still dark, and the standoff between Washington and one of Silicon Valley’s most prominent AI labs shows no sign of resolution.
What Happened on June 12
The Commerce Department issued an export control directive to Anthropic at 5:21 PM Eastern on June 12, as Anthropic detailed in its public statement. The order required the company to suspend all access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States, including Anthropic’s own foreign-born employees. Within hours, the models went dark for users worldwide.
The stated trigger was a jailbreak. Government officials told Anthropic they had identified a method for bypassing Fable 5’s safety guardrails, one that reportedly involved asking the model to read a specific codebase and fix software flaws. According to Tom’s Hardware, David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, claimed Anthropic “refused” to fix the vulnerability before the government stepped in. A Chinese group had reportedly accessed the model through the jailbreak.
Because Anthropic could not reliably verify the nationality of every user making a request in real time, it complied completely: both models were switched off for all customers, Americans included, across its entire user base. The company had no realistic alternative. You can’t selectively enforce an export ban on software that anyone with an API key can access from any IP address on the planet.
Anthropic Pushes Back
Anthropic is not taking this quietly. The company has publicly stated that the government provided only “verbal evidence of a potential narrow, non-universal jailbreak” and disagrees that this should justify recalling a commercial model deployed to hundreds of millions of users. The National Law Review’s analysis of the directive noted that the legal framework underpinning the order raises novel questions about how far export control authorities can extend into commercial AI products.
Anthropic’s core argument is that applying this standard across the industry “would essentially halt all new model deployments for all frontier model providers.” That’s a striking position. Anthropic isn’t some scrappy startup thumbing its nose at regulators; this is a company that has built its entire brand on being the responsible AI lab, the one that talks about AI safety before investors even ask. When the safety-first lab pushes back on a government safety order, it tells you something about how extraordinary the directive is.
The Bigger Picture
This is the first time the US government has used export control authorities to pull a commercial AI model off the market. The move treats frontier AI capabilities the same way Washington has historically treated nuclear materials and advanced weapons systems, a comparison that would have sounded hyperbolic two years ago but now feels uncomfortably precise.
The implications ripple outward. Every company building on top of Anthropic’s API had to scramble for alternatives when the models went dark. Enterprise customers who had built workflows around Fable 5 and Mythos 5 found themselves cut off without warning. As Snyk’s security team noted, the suspension is a wake-up call for any organization with single-provider dependency on a frontier AI model. MarketWise’s investor analysis went further, arguing that the ban introduces a new category of regulatory risk that AI investors haven’t priced in.
Meanwhile, in-person talks between Anthropic engineers and Commerce Department officials in Washington continue. Sacks has said restoration hinges solely on Anthropic patching the vulnerability. Anthropic maintains the vulnerability isn’t serious enough to warrant the recall.
Why This Matters Beyond Tech
Strip away the jargon and the dynamic here is familiar: a government asserting national security authority to control a technology, and a private company arguing the cure is worse than the disease. The Anthropic enterprise story that LNC covered last month painted a company ascendant in the corporate AI race. Now that same company is locked in a standoff with the federal government that could redefine how frontier AI gets regulated in the United States.
The precedent matters more than the specific models. If the government can pull an AI model based on a narrow jailbreak claim backed by verbal evidence alone, what stops it from doing the same to OpenAI, Google, or Meta the next time a vulnerability surfaces? And if companies know that’s the risk, will they slow down deployment, speed up patching, or simply move their most advanced research offshore?
Those aren’t hypothetical questions anymore. Day nine of the Fable 5 ban is day one of a new regulatory reality for the entire AI industry.