AI Giants Hit Pause: Anthropic Stops New Tool After US Security Order

Sat Jun 13 2026 18:12:55 GMT+0000 (Coordinated Universal Time)

Anthropic pulls its newest AI model after U.S. officials flag potential security risks, shutting down the powerful system overnight.

The Chicago‑based AI start‑up Anthropic abruptly suspended its Claude Fable 5 and Mythos 5 after a U.S. government order warned that the tools could be exploited to bypass security measures. The order came hours after the company’s public launch, leading to a sudden shutdown in order to comply with national‑security concerns. While the agency didn’t name specific weaknesses, it cited a jailbreak technique that could let users override safeguards. The move has drawn criticism that it may hamper AI safety research and collaboration with governments, and comes amid a separate lawsuit where Anthropic is fighting a Pentagon directive that labels it a security risk.

Anthropic, a Chicago‑based AI startup, pulled its new model, Claude Fable 5, from service after U.S. government officials raised concerns that the tool could be used to break security safeguards.

The order ordered the company to disable the model for all users, effectively shutting it down just days after the public launch.

Jailbreak is a method of bypassing software restrictions. The U.S. agency said it had spotted a way to “jailbreak” Fable 5, potentially letting users access hidden or dangerous functions.

Anthropic said the tool could exploit or hack computer systems, a claim that the company had highlighted before the release. It had warned that the model was “too powerful for public” because of its advanced capabilities.

The pause follows a separate lawsuit in which Anthropic is challenging a Pentagon order that banned its tech from government use, labeling it a supply‑chain risk. A U.S. judge has so far ruled the directive cannot be enforced, so the company can still work with defense agencies for the time being.

Experts warn that stopping the model could slow the pace of AI research and limit collaboration with governments. AI‑safety researcher Gina Neff says the move is “punishing research that could help keep AI systems secure.”