Anthropic Pulls Claude Fable 5 and Mythos 5 After Federal Suspension Order
A late-Friday directive citing national security forced Anthropic to cut off its top-tier models — for everyone, not just foreign nationals.
Anthropic killed access to its two flagship models on Friday evening, hours after receiving a federal order to block foreign nationals from using them.
The company said the directive landed at 5:21 p.m. ET. Rather than build a citizenship-gated access tier on a weekend timeline, Anthropic chose the blunt instrument: shut Fable 5 and Mythos 5 off for everyone, U.S. citizens included. The company called the move an "abrupt disable."
The order targets foreign nationals inside and outside the United States. That is a notable scope. It does not just apply to overseas API traffic — it reaches resident aliens, visa holders, and dual citizens physically located on U.S. soil. Enforcing that distinction at the auth layer is harder than it sounds.
This is where identity proofing collides with export control.
Most AI platforms today authenticate users with email plus password, or federated OIDC through Google and Microsoft. Neither flow surfaces nationality. Citizenship is not a claim in any standard OIDC scope, and it is not something a SAML IdP at a typical enterprise asserts in its assertion. To comply, a provider would need to bolt on a KYC-style identity verification step — passport scan, government-ID match, or a third-party verifier returning a nationality attribute. That is a substantial re-architecture of the signup and session model.
MFA would not have helped here. This is an authorization problem, not an authentication one. The question is not whether the user is who they claim to be. The question is whether that verified person is legally permitted to invoke the model. Different problem, different control plane.
There is also the session question. Existing refresh tokens for Fable 5 and Mythos 5 had to be revoked en masse. Token rotation patterns described in RFC 6749 and the OAuth 2.0 Security Best Current Practice (RFC 9700) assume you can invalidate a refresh token server-side and force re-auth. Anthropic's choice to disable the underlying model endpoint achieves the same result more bluntly: every outstanding token now hits a dead route.
Anthropic has not published a timeline for reinstatement, nor described what gated access for U.S. persons might look like. A tiered identity-verified offering seems the obvious path. Expect something resembling the export-control attestations that cloud providers already use for restricted regions.
The broader pattern is worth flagging. Frontier-model access is drifting toward the same regulatory shape as cryptography exports in the 1990s. Identity teams at AI companies should start treating nationality and residency as first-class authorization attributes now, because the next order will not come with a weekend's notice.
For the record: this is the first time a U.S. agency has ordered a domestic AI lab to cut off model access on national-security grounds. It will not be the last.
