Anthropic AI hit by two U.S. bans over jailbreak risks and export limits

The U.S. government has barred Anthropic's flagship AI models twice this year. In February the company was placed on the Department of Defense supply‑chain blacklist, prohibiting any federal agency from using its technology. A second action in May ordered a halt to foreign access to its newest models, Paeibl‑5 and Mitos‑5, citing concerns that the systems could be "jail‑broken" for illicit purposes such as bioweapon development and could be misused by China.

Anthropic CEO Dario Amodei warned that the AI could experience pain‑like sensations and pose an existential threat, a stance that drew criticism from the Trump administration, which called the firm "a radical leftist company that doesn’t understand the real world, endangering our military and threatening national security." After an audience with President Emmanuel Macron at the G7 AI summit in France, Amodei received a more positive reception. Negotiations to lift the export restrictions are reportedly under way, and the company announced the opening of its Korean office in Seoul, underscoring growing calls for "sovereign AI" capabilities that are not dependent on U.S. policy.

The episode has sparked wider debate on AI governance, export control, and the need for democratic nations to cooperate on standards, with other AI leaders such as OpenAI’s Sam Altman also urging the creation of internationally recognised rules.