[UPDATE] Mythos 5 Returns for a Trusted Few as Europe Comes Courting

The standoff over Anthropic's most powerful AI model shifted twice in a week. On June 26 the US Commerce Department revised its license terms to let a defined set of vetted organisations use Claude Mythos 5, the model Washington had pulled from general use on June 12 over security concerns. Two days later, reporting suggested Europe is trying to pull the company in a different direction.

Commerce Secretary Howard Lutnick set out the change in writing, saying he had determined that "appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model." The approved list runs to roughly 100 named US companies and federal agencies. It also covers the non-American employees at those organisations, and Anthropic's own non-American staff, who had been shut out under the original order.

The reopening comes with strings. Lutnick noted he can amend the list of partners "at any time," and reserved the right to "reevaluate and adjust the scope of license requirements on the Covered Models, should circumstances change." Anthropic reached the arrangement after working with the government to address the risks that prompted the June restrictions, which hit both Mythos 5 and the consumer-facing Fable 5. Readers who have followed the earlier turns of this story will recognise the pattern: access granted narrowly, and revocable at will.

Europe makes a play

While Washington decides who inside America may use the model, Austria is looking abroad. According to a Bloomberg report on June 28, Vienna is pressing the European Union to consider hosting Anthropic's systems on European soil, a move framed as a response to US efforts to block foreigners from reaching the company's strongest models. Reuters noted it could not immediately verify the report, and no named officials or firm details have been attached to it yet.

The idea is striking even in outline. It would treat a leading American AI model as something Europe wants to secure for itself rather than simply buy access to, much as the G7 has been searching for a way around American control of the frontier. Whether Anthropic could or would host its model weights in Europe under such an arrangement is unclear, and US export rules would have a great deal to say about it.

For now the state of play is plain enough. Mythos 5 is back, for a short and government-approved list, on terms that can change without notice. And the question of who gets to use the most capable AI has stopped being a commercial matter and become a question of statecraft.