The Trump administration's directive to suspend Anthropic's most powerful AI models, Mythos 5 and Fable 5, has ignited a fierce debate over national security, innovation, and America's role in the global AI race.
A directive from the Trump administration has forced Anthropic to pull the plug on its most powerful new AI models, Claude Mythos 5 and Fable 5. The move, citing national security fears over a potential “jailbreak,” pits one of AI’s leading labs against the full force of the U.S. government.
The order has frozen Anthropic’s flagship product and sent a chill through an industry already navigating complex issues of power, safety, and global competition.
It all started last Friday at 5:21 PM. Anthropic received a U.S. export control directive banning access to its new models for "any foreign national," a group that includes some of its own employees. This forced a complete shutdown of products that had launched just a week earlier.
The order came after a tense 90-minute ultimatum from the administration, leading to direct calls between Anthropic CEO Dario Amodei and top officials like Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick. Now, Anthropic is in Washington D.C., fighting to reverse a decision that could alter the course of American AI.
Mythos 5 is Anthropic’s apex AI, its access restricted to a handpicked list of government agencies and vetted corporate partners. The model excels at finding and exploiting cybersecurity flaws—a powerful defensive tool that becomes a significant threat in the wrong hands.
Anthropic’s own past warnings about the underlying Mythos architecture have now given the administration the leverage for its crackdown.
Fable 5 was built to be the public version of Mythos 5, wrapped in extra safeguards to make it "safe for general use." But despite these efforts, a report of a potential jailbreak—a way to bypass its safety controls—appears to have triggered the government's intervention.
Anthropic insists the vulnerability is "narrow, non-universal" and exposes capabilities already found in rival models like OpenAI's GPT-5.5.
The government’s move from product launch to mandatory shutdown was startlingly fast. A source familiar with the talks said the Trump administration gave Anthropic just 90 minutes to voluntarily restrict the models or face an official Commerce Department ban.
The ultimatum was prompted by a report, allegedly from security researchers at Amazon, which detailed a way to circumvent Fable 5’s safety features. That Amazon research, a source confirmed, was explicitly cited in conversations between Anthropic and U.S. officials.
Even as some independent security testers praised Fable 5’s protections, the government acted on the new intelligence.
Anthropic says it operated in good faith. The company pre-briefed the administration on Fable 5 and even allowed the Commerce Department to conduct pre-deployment tests, which apparently raised no alarms.
The Trump administration’s decision has provoked both fear and frustration across the tech world. Many see it as a self-inflicted wound in the global AI race with China.
A public letter, organized by Corridor CPO Alex Stamos and signed by dozens of tech and cybersecurity leaders, demanded the immediate reversal of the Fable 5 restrictions. The letter insists that AI regulation must be built on scientific evaluation and industry input, not sudden decrees.
The political uncertainty over the past weekend, he added, is already pushing companies to sign backup deals with non-U.S. AI providers.
The order sets a troubling precedent. Competitors like OpenAI, Google, and Microsoft have all released models with similar capabilities, making parallel claims about their power and potential risks.
If the administration’s rationale for banning Fable 5 stands, it could just as easily be applied to any other advanced model. This could force the industry to rally behind Anthropic or create a schism where rivals position their own models as safer, more government-friendly alternatives.
The Anthropic order marks a sharp turn in the Trump administration's AI policy. After a period of light-touch regulation, the government is now intervening directly, torn between its goal of beating China and its fear of powerful AI going rogue.
Ben Van Roo, CEO of Legion Intelligence, calls the situation "uncharted territory." He points to the near-impossibility of enforcing the directive's core demand—to block use by any "foreign national"—labeling it "the most impossible thing to enforce."
The incident also feeds a growing "AI populism," a public skepticism toward the power concentrated in a few elite labs. Van Roo warns the government’s move could stoke "greater fears and concerns, potentially for the wrong reasons."
Anthropic's team, including head of safeguards Dave Orr and researcher Nicholas Carlini, remains in Washington. Talks are described as constructive, with some officials reportedly admitting that export controls are not the right tool for the job. But with Anthropic's history of friction with the Pentagon, the path forward is uncertain.
This fight is about more than one model's security flaw. It is a defining moment that will shape the relationship between AI innovators and the government for years to come.
