Anthropic has suspended access to its newest AI model after US authorities raised security concerns only days after its public release. The company said it received an order requiring it to block foreign nationals from using Claude Fable 5, which Anthropic had described as “too powerful,” and that complying meant disabling Fable 5 and Mythos 5 for all customers.
According to Anthropic, US national security officials did not identify a specific threat, but said they had learned of a method for bypassing the model’s safety controls, often called “jailbreaking.” Anthropic said it reviewed a demonstration of the technique and found it exposed only a small number of previously known, minor vulnerabilities. The company added that similar issues could reportedly be found in other publicly available AI models without needing the same bypass.
Claude Fable 5 is part of Anthropic’s Claude family and competes with models such as OpenAI’s ChatGPT and Google’s Gemini. Before its release, Anthropic promoted the model’s safeguards against cyber abuse and said it had given early access to a limited number of organizations for testing because of concerns that the system could be used to exploit or hack computer systems. The company had warned that the model was powerful enough to be dangerous if misused.
The suspension comes amid broader tensions between Anthropic and the Trump administration. The company is already involved in a separate lawsuit over a government order that attempted to stop federal agencies from using Anthropic’s AI tools. Anthropic has also recently come under criticism from Donald Trump and from US Defence Secretary Pete Hegseth, who labeled the company a “supply chain risk,” a designation usually reserved for firms considered insecure for government use and historically associated with companies from adversarial countries.
Anthropic is challenging that designation in court. A US judge has already ruled that the Pentagon’s directive cannot currently be enforced, allowing government agencies and organisations working with the US military to continue using Anthropic’s tools while the case moves forward.
The company said the government’s concern appears to stem from a bypass technique rather than a specific confirmed breach, and insisted the issues discovered were minor. Still, the abrupt suspension highlights growing pressure on AI firms to balance rapid model development with national security, cybersecurity, and regulatory scrutiny. As AI systems become more capable, governments are paying closer attention to how they can be restricted, tested, and potentially abused.
