The U.S. export‑control office issued a directive Thursday that blocks foreign nationals—including Anthropic staff—from accessing the company’s newest language models, Fable 5 and Mythos 5. Anthropic responded by pulling both models from all servers worldwide within hours. The ban applies to the U.K., Canada, Australia, New Zealand and any of their citizens, regardless of residence, and any affiliated research labs.

Because Fable 5 launched only three days ago and Mythos 5 remains limited to select partners, immediate user disruption is minimal. However, this marks the first export‑control order targeting LLM access, signaling that any U.S.–based model could vanish without warning immediately. Isaacus, an Australian legal‑AI startup, cites the move as proof that frontier AI must remain self‑hostable and independent of proprietary clouds.

Isaacus has kept every released model available for air‑gapped self‑hosting since day one and vows to double down on AI sovereignty. The strategy lets small teams run legal‑focused models on consumer hardware, securing contracts with multiple Australian government agencies. Upcoming products like the Blackstone Graph and Isaacus Research platform will also ship fully self‑hostable, strategically ensuring continuity regardless of external policy shifts.