U.S. Export Directive Forces Anthropic Model Access Cut

The U.S. Department of Commerce sent a directive to Anthropic on June 12, 2026, citing national security authorities, ordering the suspension of all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including Anthropic's own non-citizen employees, per Anthropic's official statement. Anthropic said the net effect required disabling both models for all customers to ensure compliance. Access to all other Claude models was not affected.

Axios and the Wall Street Journal report that Amazon CEO Andy Jassy personally escalated a reported jailbreak technique to Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and National Cyber Director Harry Coker Jr. The reported jailbreak involved a "fix this code" prompt that researchers found could coax Fable 5 to surface cybersecurity vulnerability information. That evening, Lutnick sent the enforcement letter to Anthropic CEO Dario Amodei. The letter did not provide specific details of its national security concern, per Anthropic's statement.

Anthropic's official statement contested the rationale. The company said it reviewed the government's demonstration and found the jailbreak narrow and non-universal - essentially limited to asking the model to find software flaws in a specific codebase. Anthropic stated that comparable capability is available from GPT-5.5 and other publicly deployed models, and that no universal jailbreak had been found. The company argued that a defense-in-depth strategy, including 30-day data-retention requirements on Mythos-class models, was designed to mitigate exactly this class of risk, and that applying the government's standard across the industry would "essentially halt all new model deployments."

TechCrunch reported that the ban was likely "reactionary" or "retaliatory" rather than primarily technical. Axios reported a sustained failure of communication between Anthropic and the Trump administration, quoting an administration official as saying "They screwed us," with sources describing Anthropic as unable to bridge ideological differences with the current administration. Prior reporting by Politico (April 2026) and CNBC (May 2026) noted Anthropic had already been in conflict with the administration over Pentagon contract terms and a supply-chain-risk blacklisting.

Axios and Forbes reported that Anthropic met with Commerce Department officials and the White House to seek restoration of model access. AP reported that cybersecurity experts urged the White House to ease restrictions. Material signals to watch: any public release of the enforcement letter, a technical disclosure of the alleged jailbreak mechanics, and whether the administration codifies or relaxes the export-control criteria applied here.

The directive set a novel precedent - a non-public government enforcement action immediately suspended two leading hosted AI models for all users globally. The combination of contested technical rationale, reported personality clashes, and building security-community pushback creates significant uncertainty for organizations assessing cross-border access and vendor risk for advanced AI models.