Anthropic has halted the rollout of its cutting‑edge AI model, Claude Fable 5, after U.S. government officials raised concerns about potential security risks.
In a statement posted on its website, the company said it was “ordered to suspend foreign nationals from using Claude Fable 5,” a program it self‑described as “too powerful.” The order required Anthropic to abruptly disable Fable 5 and Mythos 5 for all customers to ensure compliance.
Anthropic alleges the U.S. national‑security authorities have not identified specific concerns but believe the government has learned of a method to bypass or “jailbreak” Fable 5. Jailbreaking is the process of circumventing software safeguards that keep a network secure, potentially exposing sensitive data.
The firm said it reviewed a demonstration of a jailbreak technique that exposed a small number of known, minor vulnerabilities, but noted similar flaws were found in publicly available models without any enforcement of a jailbreak.
This pause follows earlier warnings from finance, technology and government leaders who voiced concerns about the public release of the model, which had been privately tested in April to probe internal security weaknesses.
Anthropic’s CEO had cautioned that Fable 5 was “too powerful to release” prior to its public launch, a claim that critics say is marketing hype. Nevertheless, the company insists the model’s capabilities surpass those of any prior publicly available AI.
The U.S. government has also taken a more direct stance: President Trump publicly criticised the company and Secretary of Defense Pete Hegseth labeled it a “supply chain risk,” the first time a U.S. company received such a designation. A U.S. judge ruled the Pentagon’s directive could not be enforced, allowing government agencies to continue using Anthropic tools while the lawsuit proceeds.
The clash underscores a growing push-and‑pull between AI innovation and national security, with Anthropic’s case now a high‑profile example of how the federal government is tightening scrutiny over AI services that could potentially pose strategic risks.
