It has been nearly a week since the Trump administration issued an export control directive to Anthropic, leaving one of the world’s most prominent AI labs in a state of limbo. The directive forced the company to pull its most advanced models, Claude Mythos and Fable 5, offline entirely. Negotiations between the company and the White House are ongoing, but the two sides remain at odds over how to resolve the situation and restore the tools to the public.
As reported by The Wired, the dispute is defined largely by its lack of transparency. At no point has the United States government formally stated what Anthropic did wrong. The clearest insight into the administration’s position came through a post on X by White House technology adviser David Sacks, who said officials acted reluctantly after Anthropic CEO Dario Amodei allegedly refused to fix a jailbreak or de-deploy the model.
Anthropic’s account of events differs. According to a person close to the company, Anthropic does not believe it violated any concrete procedures or rules laid out by the Trump administration. With few laws currently governing frontier AI development, companies have found themselves facing penalties for crossing lines that were never clearly defined.
The shutdown’s consequences stretched well beyond Anthropic’s own teams
The administration demanded that Anthropic prohibit all foreign nationals from accessing Mythos and Fable 5. Because the order applied to foreign national employees at Anthropic itself, the company disabled the models for its internal research teams as well as its entire customer base, which includes major partners like Meta and Apple.
The White House has expressed concern that Anthropic shared Mythos with SK Telecom, a South Korean firm officials allege has ties to China. Amazon CEO Andy Jassy reportedly raised separate concerns to Treasury Secretary Scott Bessent that the guardrails on Claude Fable 5 could be circumvented.
Anthropic has pushed back on both points, saying it coordinated with the government on the Mythos rollout and revoked access to the South Korean partner immediately after officials raised concerns. The episode follows a separate clash in which the Pentagon barred military contractors from working with the company entirely.
The jailbreaking concern adds further complexity. Anthropic and independent cybersecurity researchers have long argued that eliminating jailbreaks entirely is not a simple or isolated task, since AI models are probabilistic rather than deterministic and cannot guarantee a fixed output for every prompt.
Anthropic said it reviewed a demonstration of a narrow, non-universal jailbreak that involved asking the model to read a codebase and fix software flaws. It further argued that recalling a commercial model over such a finding would, if applied industrywide, halt new model deployments for every frontier provider.
This standoff reflects a broader shift in how the Trump administration approaches AI oversight. President Trump signed an executive order creating a voluntary system for labs to submit models for early testing, with language explicitly meant to prevent it from becoming a mandatory licensing regime.
A former White House technology official said the administration should not have called the system voluntary, since it now functions as a de facto licensing regime. The administration had previously signaled a lighter regulatory touch before tightening its posture in this case.
Other labs, including Google, OpenAI and Meta, are watching the situation closely. Many industry leaders have concluded that offering the White House early access to upcoming models and being proactive about sharing launch details is now the safest path forward. Cohere CEO Aidan Gomez said in an interview that advance notice and advance access have become the primary requests from authorities.
For now, users remain locked out of Fable 5 and Mythos, with Anthropic and the government still working toward an agreement on modified safeguards.
Published: Jun 18, 2026 08:30 pm