The reversal hands more than 100 US institutions access to Anthropic's most powerful model, but leaves Anthropic's sibling model Fable 5 in limbo and reads as a negotiated settlement rather than a safety capitulation.
Two weeks after using export controls to halt Anthropic's most powerful AI model, the Trump administration quietly reversed course in a Friday letter to the company. Claude Mythos 5, the system's public codename, will now reach more than 100 US institutions, including major companies and government agencies. The reversal reads less like a safety capitulation and more like a negotiated settlement: access traded for institutional commitments, with the smaller sibling model Fable 5 left conspicuously outside the deal.
The original block, reported by Semafor and surfaced on the Techmeme wire, came after warnings, most prominently from Amazon, that Mythos 5 and Fable 5 could be jailbroken for malicious purposes. Anthropic's own statement on the prior suspension acknowledged the US government's directive without contesting it. Both models were pulled from circulation. Fable 5 had, briefly, been the most capable AI system widely available to consumers before the pause, and its absence registered quickly in the developer community within hours of the letter.
What changed in two weeks is the more interesting story than the original block. Export controls on a frontier AI model are an unusual instrument. They are not a product recall, a safety finding, or a court order. They are a piece of state leverage aimed at a private lab, deployed because regulators believed a powerful system could cause downstream harm if widely released. The decision to lift that leverage after a fortnight, without public hearings, technical adjudication, or a documented change in the underlying jailbreak risk, suggests the instrument was always intended to be tradable.
That framing has real consequences for how the rest of the AI industry should read this episode. The Friday letter, reported by Semafor via Yahoo News, is silent on Fable 5, and people close to the talks say Fable's release is being negotiated separately with no clear timeline. The asymmetry is informative: Mythos 5 is the larger lever, and Fable 5 is the residual bargaining chip. By reopening Mythos while leaving Fable frozen, the administration preserves a position to trade from in a second round.
It is worth being precise about what is and is not known. The "more than 100 US institutions" figure comes from a single Semafor report via Yahoo News and should be treated as a reported count rather than a confirmed headcount. The exact letter, the agency that signed it, and the institutional criteria for inclusion have not been published; Anthropic's product page for the Claude Fable 5 and Mythos 5 family describes the technical systems but not the access regime. The Wikipedia entry on Claude Mythos is background only and should not be treated as the primary record of the Friday decision.
The watch item, then, is not whether Mythos 5 ships. It presumably will. The watch item is whether the administration treats this template as repeatable: a short, opaque export-control action used to extract terms from a frontier lab, then unwound in exchange for institutional access and a still-frozen second model. If Fable 5 is released on a similar letter a few weeks from now, the playbook is confirmed. If Fable 5 stays frozen indefinitely, the leverage simply shifts to the next round of negotiation over the next model Anthropic tries to ship.