On June 9, 2026, Anthropic launched Claude Fable 5 and Claude Mythos 5. Fable 5 scored 95 percent on SWE-bench Verified. Mythos 5 scored 95.5. By any available measure, these were the most capable coding models Anthropic had ever shipped, and for the three days they were available, the response was what you would expect: teams started migrating workflows, developers started testing, the early reports came in positive.
On June 12, both models went offline. Not because of a technical failure. Not because Anthropic made a business decision. Because the Commerce Department's Bureau of Industry and Security, acting under the authority of national security statutes and with the signature of Commerce Secretary Howard Lutnick, issued an emergency export control directive requiring Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national — whether inside or outside the United States, including Anthropic's own employees who are not US citizens.
Anthropic's public statement was precise about the problem: the company cannot reliably determine which of its users is a foreign national in real time across a user base of hundreds of millions. The directive left no room for partial compliance. The result was a hard global shutoff for all customers.
The stated rationale involves two separate threads. The first is a data breach: a China-linked group had reportedly accessed Mythos using credentials obtained in an earlier intrusion, sometime in April. The second is a jailbreak: Amazon researchers discovered that a multi-step prompting sequence could extract from the model detailed information about cybersecurity vulnerabilities associated with critical infrastructure — the kind of capability that, in the wrong hands, has obvious offensive applications. Amazon CEO Andy Jassy personally raised concerns with senior White House officials after his researchers demonstrated the technique.
Anthropic's position is that the bypass is narrow and non-universal — that it amounts to asking a model to read a codebase and identify software flaws, which is a thing other publicly available frontier models, including OpenAI's GPT-5.5, can do when prompted appropriately. The Trump administration's AI adviser David Sacks reportedly said Anthropic was warned about the jailbreak before the export order and declined to pull the model voluntarily. Anthropic disputes the characterization that there was a viable fix available and implemented. As of June 22, both models remain offline. Anthropic has said it believes the situation is a misunderstanding and is working toward restoration. No restoration date has been confirmed.
What makes this different from anything the US government has done to an AI company before is the instrument. The United States has used export controls on semiconductors — the H100 chip export restrictions, the A100 restrictions, the successive tightening of the Entity List rules around Nvidia hardware. These controls work because chips are physical objects that travel across borders. Applying export control law to a software API is structurally different. An API does not cross a border; it is accessed from wherever the user sits. Blocking foreign nationals from accessing an API means either building nationality-verification infrastructure that does not exist at commercial scale, or doing what Anthropic did: turning it off for everyone.
The National Law Review's analysis notes that this is the first time a government has used export control authority to suspend access to a deployed AI model API, and that the legal mechanism is being improvised in real time. The Bureau of Industry and Security was built to regulate physical goods and hardware. Its tools for software services are less developed, and the directive itself reflects that: it told Anthropic what it could not allow, without specifying how to achieve compliance short of a global shutoff.
For builders, the practical consequence has nothing to do with export law. It is simpler: a frontier model you depend on can be removed from availability with no warning, for reasons outside the model provider's control, with no recourse and no timeline for restoration. Teams that had moved agentic coding workflows onto Fable 5 in the three days it was available had to roll those workflows back to Opus 4.8 or GPT-5.4 or reroute entirely. The productivity loss is real; the business continuity question is real.
Anthropic has published its full statement and it is worth reading in full, not because it resolves anything — it does not — but because it is a technically careful account of an impossible compliance situation from a company that appears genuinely frustrated by the order.
The sector has been watching this play out and, for the most part, quietly. Other labs have models with similar capability profiles. OpenAI has not received a similar directive, despite Anthropic's public argument that GPT-5.5 can produce the same result through the same technique. Whether the government's action reflects a specific judgment about Fable 5 and Mythos 5, a specific concern about Anthropic's compliance posture, or the first move in a broader wave of model-level export controls is not yet clear from anything publicly available.
The US government found the off-switch for a frontier AI model, and nobody — not the company, not the users, not the legal system — has a clear map of what comes next.
The short of it.
On June 12, the Commerce Department ordered Anthropic to suspend all access to its two newest and most capable models — Fable 5 and Mythos 5 — for foreign nationals, citing a China-linked access incident and a jailbreak technique that Amazon researchers demonstrated to White House officials. Because Anthropic cannot verify user nationality in real time, both models went offline for all users globally. As of June 22, they remain offline. This is the first time US export control law has been used to pull a deployed AI model API, and it establishes a precedent that any frontier model can be severed from its users without notice and for reasons outside the provider's control. Build your continuity planning around that fact.