The American government has, for the first time, ordered a private artificial intelligence company to deactivate its most powerful products. Anthropic's Fable and Mythos model families—including the recently launched Fable 5, which the company had positioned as its answer to OpenAI's frontier systems—are now dark, suspended by executive directive rather than corporate decision.
This is not a drill. It is not a voluntary pause for safety research, nor a commercial pivot. It is the federal government reaching into a San Francisco company's servers and flipping the off switch.
The directive and its discontents
The Trump administration's order, details of which remain partially classified, reportedly cites national security concerns around the models' capabilities in certain sensitive domains. Amazon CEO Andy Jassy had apparently raised concerns about Anthropic's models to administration officials before the crackdown—a notable intervention given Amazon's substantial investment in the company and the competitive dynamics that creates with Amazon's own AI ambitions.
Anthropic has complied, though the company's public statements suggest something between grudging acceptance and quiet fury. The firm did manage to release a consumer-accessible version of Mythos before the shutdown, and cybersecurity researchers have already begun cataloguing the guardrails imposed on Fable 5 in its brief public life—guardrails they largely view as excessive and counterproductive.
The list of topics Anthropic says are "too dangerous" for Fable 5 to discuss has itself become a subject of intense debate, with critics arguing the restrictions reveal more about political sensitivities than genuine safety concerns.
The constitutional question nobody wants to answer
The legal basis for this intervention is murky at best. The administration appears to be operating under emergency national security authorities, but no court has reviewed whether those authorities extend to commanding a company to disable commercial software products already in customer hands. First Amendment scholars are quietly alarmed; this is prior restraint on a scale previously confined to law school hypotheticals.
The practical implications compound the legal ones. Anthropic's enterprise customers—hospitals, law firms, financial institutions—now face the sudden disappearance of tools they had integrated into critical workflows. The company's India operations have become a particular flashpoint, with the Indian government now debating whether American regulatory overreach should accelerate its own domestic AI development.
Our take
The Fable and Mythos models may well have posed genuine risks; frontier AI systems are not toys, and reasonable people can disagree about acceptable capability thresholds. But the mechanism matters as much as the outcome. A government that can order AI models deactivated by fiat can order them activated by fiat, or modified by fiat, or deployed for surveillance by fiat. The Trump administration has just established that American AI companies operate at the pleasure of the executive branch. Every competitor, every ally, every adversary is watching—and drawing conclusions about where to build their next data center.
