U.S. export controls cut global access to Anthropic models

Anthropic disabled its two most powerful models after receiving a U.S. export-control letter directing suspension of access by any foreign national. The company said the order arrived at 5:21 p.m. ET on June 12 and that, lacking a real-time way to filter users by nationality, it had to “abruptly disable” access for all customers to comply.

Claude Fable 5 was made generally available on June 9. Mythos 5 remained under a restricted programme and was available only to selected partners and researchers. U.S. officials cited a national security concern tied to a method for bypassing Fable 5’s safety guardrails. Anthropic disputed the government’s assessment, saying the technique mainly allowed limited review of program code and that other models can perform similar tasks.

David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, wrote that the administration asked Anthropic to fix the reported vulnerability or remove the model and that company leadership declined. He added, “The ball is in Anthropic’s court.” Officials have noted that some researchers used model prompts in ways that could provide information relevant to cyberattacks.

European officials and lawmakers reported immediate effects on businesses and research institutions. The European Commission said the new class of models offers benefits including for cyber-defence but raises cybersecurity concerns requiring non-discriminatory contingency measures. A Finnish member of the European Parliament warned Europe cannot continue to rely on access that can be turned off by a foreign government overnight. Canada’s prime minister described the episode as an example of the risks of depending on a small number of U.S. providers and called for diversification. Britain’s minister for AI and online safety urged more domestic investment in AI capacity.

The Commerce Department’s Bureau of Industry and Security placed Fable 5 and Mythos 5 under export controls that require licences for export, re-export or domestic transfer. Reinstatement of access for foreign nationals will depend on individually validated licences and any technical fixes Anthropic implements. The department has the authority to impose civil penalties for non-compliance.

Anthropic has warned that applying the same licensing standard across the industry would halt new model deployments for frontier providers. The company has also contested previous U.S. determinations that limited federal use of its technology; earlier this year the Pentagon designated the firm a supply-chain risk, and Anthropic filed suit to reverse that designation, saying it could affect substantial revenue.

Governments are reviewing the incident and considering policy responses. The European Commission published a package on technological sovereignty, including a draft Cloud and AI Development Act, shortly before the export controls took effect. For now, companies and researchers outside the United States await licence decisions and any updates Anthropic makes to address the reported vulnerability.