The US government ordered Anthropic on a Friday evening to immediately block access to its two most advanced AI models — Claude Fable 5 and Claude Mythos 5 — for all foreign nationals anywhere in the world, according to The Verge. Anthropic complied by cutting off the models entirely for every customer globally, not just overseas users.
The ban was sweeping enough to include foreign nationals working inside the United States — even Anthropic's own employees, according to Tom's Hardware. All other Anthropic models, including Opus 4.8, remain available.
The government cited national security concerns, but the precise trigger appears to be a jailbreak report by Amazon researchers, according to Luta Security's CEO, who told the Wall Street Journal the restrictions amount to a "complete overreaction." Anthropic pushed back publicly, saying the vulnerabilities cited are minor and exist in competing models including GPT-5.5, according to The Decoder.
Anthropologic called the order a "misunderstanding," according to reporting by MSN. The company had separately declined to publicly release Claude Mythos after internal testing reportedly showed the AI could autonomously discover and exploit serious cybersecurity vulnerabilities, according to MSN.
The episode carries an uncomfortable irony noted by TechCrunch: Anthropic built its brand around transparent AI safety warnings, and those warnings may now be fueling government alarm. India, which accounts for Claude's second-largest user base after the US, is among the hardest-hit markets, according to Hindustan Times. Meanwhile, Punchbowl News reported the Pentagon has been moving to distance itself from Anthropic.
The shutdown signals that AI models have officially crossed into the territory of strategic export controls — the same category once reserved for weapons systems and semiconductor technology.