U.S. Commerce Secretary Howard Lutnick said he took action against Anthropic’s latest Mythos and Fable AI models because officials feared they could be deployed by military intelligence users in China, Russia, or other countries of concern.
In a letter sent to Anthropic CEO Dario Amodei on Friday, Lutnick ordered the company to suspend export of the models to destinations worldwide and to all foreign nationals, wherever located, according to a copy seen by Reuters. The directive cited national security authorities under the 2018 Export Control Reform Act, marking the first time the Commerce Department has invoked this power for an AI model.
Anthropic immediately disabled access to Fable 5 and Mythos 5 for all users globally to ensure compliance. Senior technical staff from the company met with Commerce Department officials in Washington on Monday to negotiate a resolution, according to a Trump administration official and a person close to the company.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
National Cyber Director Sean Cairncross joined the discussions, and talks have continued virtually daily since Friday. Lutnick has been personally involved, holding regular calls with Anthropic executives, with both he and Amodei expected to attend the G7 meetings in France, where further negotiations may occur.
The government is seeking ironclad assurances that the models cannot be used to harm U.S. interests, while Anthropic is pushing to restore access to its top-tier systems after taking them offline. The company described the dispute as a “misunderstanding” centered on a “potential narrow, non-universal jailbreak” that could allow a user to bypass safeguards and ask Fable 5 to analyze and fix software vulnerabilities. Anthropic argued the flaws identified were minor and comparable to those findable by other publicly available models.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” the company said in a Friday blog post. “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Background of Strained Relations
The latest clash builds on earlier tensions. Earlier this year, the Pentagon placed Anthropic on a national security blacklist after the company refused to allow its Claude models to be used for domestic surveillance or fully autonomous weapons. That designation banned defense contractors from using Anthropic technology. Anthropic sued to reverse the move, and the litigation is ongoing.
Despite the friction, the company had worked with government agencies to test Fable 5 and Mythos 5 before their release and received initial approval for deployment.
Cybersecurity leaders have rallied in support of Anthropic. In an open letter sent Sunday to Lutnick and Cairncross, more than 80 executives and experts, including from Nvidia and Adobe, argued that the restrictions harm defenders at a time when China-linked hackers pose the biggest espionage threat. The letter warned that removing access to powerful defensive tools risks America’s AI leadership without clear justification.
Export control experts have raised doubts about the legal basis for the directive. AI models are typically accessed remotely rather than physically exported, and current regulations may not clearly cover such deployments. The Commerce Department has not publicly detailed its authority, and the move could face legal challenges if Anthropic contests it further.
Anthropic’s safety-focused posturing, illustrated by the limiting of initial rollouts of powerful models and refusing certain military applications, has done little to cut it some slack. Its rapid commercial scaling and high valuation (nearing $965 billion with a confidential IPO filing) have intensified scrutiny.
However, besides its impact on Anthropic’s growth, Washington’s decision is setting a precedent for the AI industry. Industry analysts have warned that if other governments or regulators adopt similar standards, frontier model providers could face repeated hurdles in deployment, potentially slowing innovation and raising compliance costs.
Anthropic said it is working to address the government’s concerns and restore access “as soon as possible.” A productive meeting with government officials is expected to lead to revised safeguards or a narrower restriction, allowing the models back online with enhanced monitoring. Failure to resolve the issue quickly risks broader market uncertainty and could complicate Anthropic’s IPO preparations.
The dispute comes at a sensitive time for U.S. AI policy. The Trump administration has emphasized voluntary cooperation and pre-release testing but is also signaling a harder line on perceived risks. Against this backdrop, industry leaders are praying that the focus remains on finding common ground.