Anthropic has paused its latest AI models, Claude Fable 5 and Mythos 5, after U.S. authorities flagged security worries just days after their public release. In a statement published on its website, the company said it was ordered to suspend foreign nationals from using Claude Fable 5, a program it had described as “too powerful.” “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance,” Anthropic wrote.
The situation unfolds amid a separate legal dispute between Anthropic and the Trump administration over an order that would bar government agencies from using the company’s tools. The BBC has sought comment from the U.S. Department of Commerce on the matter. Claude Fable 5 is a version of Anthropic’s Claude Mythos, a system positioned to compete with OpenAI’s ChatGPT and Google’s Gemini.
Anthropic said U.S. national security authorities had not identified specific concerns, but that the government believes it has become aware of a method to bypass, or “jailbreak,” Fable 5. The company said it had reviewed a demonstration of this technique, which appeared to reveal a small number of previously known, minor vulnerabilities. Anthropic added that these vulnerabilities appear relatively simple and that other publicly available models can discover them without a bypass.
Ahead of Fable 5’s release, Anthropic claimed to have implemented safeguards to counter cyber hacking. The firm had granted pre-release access to a limited set of organisations, arguing the tool’s capabilities were powerful enough to pose risks if misused.
The company said, “Fable’s capabilities exceed those of any model we’ve ever made generally available.” In June, the European Union — which had gained access to Mythos after talks — said the development underscored Europe’s push for technological sovereignty. A Commission spokesman said: “We take note of Anthropic’s statement and are assessing” the bloc’s dependence on non-European technology for AI.
Industry voices have warned that restricting access could hamper the safe testing and development of AI systems and may limit international collaboration. A Queen Mary University of London professor told the BBC that the move could curb global cooperation and the safe testing of AI tools, noting that the AI Security Institute in the U.K. found the model could exploit defences in tests.
In political dynamics, former U.S. President Donald Trump has criticized Anthropic publicly, while U.S. Defence Secretary Pete Hegseth labeled the company a “supply chain risk” — a designation historically reserved for entities deemed not secure for government use. Anthropic is pursuing legal action against the Pentagon over that designation. A U.S. judge has ruled the Pentagon directive could not be enforced, leaving government agencies that work with the U.S. military able to use Anthropic’s tools while the case proceeds.
