Anthropic, Fable and Mythos: what the government feud means

Anthropic released a modified model called Fable to the public on Tuesday, June 9, after saying in April that it had built a model called Mythos that was so good at working with code it could pose a global cybersecurity threat. That Friday the federal government told the company it was a threat to national security and placed export controls on the new release, and Anthropic revoked access to both models hours later.

What happened and when?

Anthropic unveiled Mythos in April as an AI system highly capable with code, then shared access with a small group of cybersecurity experts so they could assess its risks; on Tuesday, June 9 the company released a modified public version called Fable. Three days later the federal government labeled the release a national-security threat and imposed export controls, and Anthropic cut access to Mythos and Fable hours after the government action.

The episode followed a pattern: Anthropic disclosed a powerful coding model, gave limited expert access, and then pushed a public-facing variant. The government reaction was swift and decisive; the company also faced outside pressure, with Amazon CEO Andy Jassy telling officials that Fable would be dangerous, while Amazon remains both an investor in Anthropic and a builder of its own competing AI models.

How could this change who companies trust for AI?

The government’s move is already pushing some organizations away from US providers, and European politicians framed the incident as a prompt to build more domestic AI capacity. French politician Bruno Retailleau called it a “wake-up call” and argued Europe should accelerate its AI efforts.

At the same time the story highlights an obvious alternative: open-source models from China, which the article describes as “very capable and incredibly cheap” and downloadable to run on anyone’s servers without rules or guardrails. That availability makes them attractive to companies that fear access could be cut off by White House decisions, and the piece notes investor appetite reflected in the “skyrocketing” shares of the Chinese startup Zhipu.

Those two pressures pull in opposite directions. Relying on foreign open-source models avoids US government shutdown risk, but it also raises the very cybersecurity concerns Anthropic said it was trying to address by building safety guardrails into its models.

Why it matters

Cutting access to Anthropic’s models risks undermining the very defenses researchers were building: leading cybersecurity experts argued in an open letter that access to Anthropic’s models was helping defenders prepare, and that those models were no more dangerous than other widely available leading models. The decision therefore creates a policy trade-off: restrict distribution to reduce one class of risk, or preserve researcher access to improve defenses against real threats. The episode also amplifies political pressure for federal AI regulation, after earlier clashes with the White House spurred bills aimed at defining limits on military AI.

What to watch

Watch whether the export controls survive legal scrutiny and how lawmakers respond: the government’s action could prompt new legislation or oversight in the coming months. Also watch for any explicit moves to treat US use of foreign models as a national-security risk, and for further shifts by companies toward Chinese open-source models or other non-US providers.