Anthropic feud tests US AI controls

Anthropic’s dispute with the US government escalated after the company said in April that Mythos, an AI model built for coding, could pose a global cybersecurity threat. It gave a small group of cybersecurity experts access, then released a modified public version called Fable on Tuesday, June 9, describing it as safer. That Friday, the federal government labeled the release a national security threat, imposed export controls, and Anthropic revoked access to both models hours later.

The move has raised doubts about relying on American AI providers, especially in Europe, where French politician Bruno Retailleau called it a “wake-up call” for more domestic AI development. The shift could also push companies toward capable, cheap open-source models from China that can run on private servers without US guardrails, a dynamic reflected in rising shares of Chinese startup Zhipu.

Cybersecurity experts warned in an open letter that cutting off Anthropic’s models could leave the country more vulnerable by limiting defensive research, especially if other widely available models offer similar capabilities. The episode also increases pressure on lawmakers to define rules for AI safety and military use, as the White House shifts between promising lighter oversight and treating Anthropic’s releases as national security risks.