Anthropic shut down two of its most powerful AI models on June 12, 2026, after the US government issued an export control directive requiring the company to block all foreign nationals from accessing Fable 5 and Mythos 5. Unable to verify citizenship in real time, Anthropic made the call to pull both models offline entirely. This appears to be the first time a major AI lab has withdrawn a publicly released model at the direct order of a federal agency.
The directive arrived at 5:21 PM on Friday, June 13, 2026, signed by Commerce Secretary Howard Lutnick and addressed to CEO Dario Amodei, according to Axios. The order invoked broad national security authorities without specifying the underlying concern. Mythos 5 had been reserved for participants in Project Glasswing, Anthropic's program built around the cybersecurity capabilities of its most advanced model. Fable 5, by contrast, had only been made publicly available days earlier. All other Anthropic models remain operational.
The Alleged Jailbreak and What Anthropic Found
Anthropic's account is that the government believed it had identified a method for bypassing Fable 5's safety filters. A source cited by Axios said the action was triggered after a separate company claimed to have unlocked Mythos. Anthropic says it reviewed a demonstration of the technique and found only a small number of minor, already-known vulnerabilities. The technical argument is pointed: the same capability is widely available on other models, including OpenAI's GPT-5.5, and is routinely used by defenders maintaining critical systems. Blocking Anthropic's models on these grounds, while equivalent tools remain publicly accessible, raises difficult questions about the consistency of the policy.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
— Anthropic (@AnthropicAI) June 13, 2026
The net effect of…
“Defense in Depth”: Anthropic's Security Argument
Functionally, anthropic is pushing back on the premise. The company says it conducted thousands of hours of red-teaming on Fable 5's filters, working alongside the US government, the UK's AI Safety Institute (AISI), and independent external teams, and that no one has yet found a universal jailbreak capable of broadly unlocking the model's most sensitive functions. Anthropic's position is that perfect jailbreak resistance does not exist for any provider, and that the right response is making attacks either extremely narrow or extremely costly, backed by continuous monitoring. For Fable 5 specifically, Anthropic had imposed a 30-day data retention requirement precisely to study and neutralize misuse. The company's full statement is published on the Anthropic website.
Why This Precedent Matters Beyond One Friday Afternoon
Anthropic has accepted the directive while contesting the principle behind it. Withdrawing a model already distributed to hundreds of millions of users over a narrow jailbreak, the company argues, would set a standard that effectively blocks every future release across the entire sector. Anthropic acknowledges that governments must retain the power to halt dangerous deployments, but insists this must happen through a transparent, fair, and technically grounded statutory process. This action, in the company's view, meets none of those criteria.
The broader stakes are significant. Export controls have historically targeted chips and hardware. Applying them directly to a software model is a different kind of intervention, and one that intersects with the regulatory tightening already underway through the EU's Digital Omnibus and the first national AI Act implementation laws. The authority behind this directive is the US Department of Commerce, the same body that oversees the export of sensitive technologies. Anthropic has apologized for the disruption, described the situation as a likely misunderstanding, and said it is working to restore access as quickly as possible. Whether the government follows up with technical evidence, or the standoff resolves quietly, will say a great deal about how federal authorities intend to regulate AI models that are already in the hands of the public.
