What Exactly Did the Lutnick Letter Demand?

According to Bloomberg Technology, which obtained the full letter, Commerce Secretary Howard Lutnick wrote to Anthropic CEO Dario Amodei on Friday June 12, 2026, asserting that the company would need "a license from the Bureau of Industry and Security (BIS) before exporting or re-exporting the Fable 5 and Mythos 5 artificial intelligence models to any destination worldwide, or to any foreign national regardless of location." The letter cited the Export Administration Regulations (EAR) and claimed that these models could be used for "weapons of mass destruction-related end uses or end users." Lutnick explicitly warned that failure to comply could result in "criminal penalties, including fines and imprisonment." This is the first time the US government has applied such sweeping, worldwide restrictions to a commercially available AI model.

Why Did Anthropic Disable Mythos 5 Instead of Seeking a License?

The answer lies in the sheer breadth of the demand. Anthropic's internal analysis, which the company shared with Bloomberg, concluded that seeking licenses for every single foreign user and foreign national — including researchers, students, and developers — would be logistically impossible and would effectively halt all international use of the model. Anthropic said in a statement that "the scope of the license requirement is unprecedented and makes continued global access to Mythos 5 untenable under current timelines." By disabling the model entirely, Anthropic avoids the risk of violating the order while buying time to negotiate a narrower framework. This decision, however, immediately angered the global developer community that had built applications on Mythos 5.

Who Loses Most From This Export Control Escalation?

The clear losers are non-US AI developers and researchers who depended on open access to frontier models. According to Anthropic's developer documentation, Mythos 5 had been downloaded over 2 million times and was used in production by more than 15,000 companies outside the United States as of June 2026. These users now face a sudden loss of capability with no clear alternative. The winners, paradoxically, may be US-based competitors like OpenAI and Google DeepMind, which are not subject to the same restrictions because their models are primarily accessed via cloud APIs rather than downloadable weights. According to a report by the Center for Security and Emerging Technology (CSET), this creates a "moat" around US AI companies that operate behind API walls, while punishing those like Anthropic that pursued a more open distribution model.

Is This a One-Off or a New Regulatory Paradigm?

The language in Lutnick's letter suggests this is not an isolated incident. The Secretary explicitly warned that "BIS may take similar actions with respect to other frontier AI models if they pose comparable national security risks." This signals that the Commerce Department now views itself as having the authority to unilaterally designate any AI model as a controlled item under the EAR. According to a former BIS official quoted by Bloomberg, "This letter is a shot across the bow for every AI lab. If you release a model that the government thinks is too powerful, they can now order you to pull it from the market." The precedent effectively turns the Commerce Secretary into a de facto AI regulator with the power to shut down model access globally.

1. Specific company, action, and timeframe: By December 2026, the US Commerce Department will issue a similar export control order targeting Mistral AI's next-generation model (the successor to Mistral Large), citing national security concerns.
2. Specific market outcome and timeframe: Within 18 months, at least three major non-US AI labs (including Baidu and a European consortium) will announce open-weight models that explicitly target developers displaced by US export controls.
3. Specific regulatory action and timeframe: By June 2027, the US Congress will introduce legislation to codify the Commerce Secretary's authority to designate AI models as controlled items, effectively ratifying the Lutnick precedent.