The US Commerce Department has issued an order requiring AI company Anthropic to restrict access to its most advanced AI models from foreign nationals. This move is seen as an attempt by the US government to maintain control over cutting-edge AI technology and limit its global distribution.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
In June 2026, the U.S. Commerce Department, citing national security, ordered Anthropic to bar all foreign nationals from accessing its most advanced AI models. Anthropic was forced to shut down the models globally and restored access only after reinforcing safety guardrails, but model performance plummeted. The incident sparks legal controversy: Do existing export control laws apply to AI models? Is the government's action lawful?
On June 12, 2026, the U.S. Commerce Department sent a letter to Anthropic ordering it to immediately prevent all foreign nationals from accessing the company's two most advanced large language models—Claude Fable 5 and Mythos 5. Anthropic shut down both models within hours, making them unavailable to everyone. The reason: the company could not instantly collect and verify the nationality of each user, including overseas users.
On June 30, the Commerce Department lifted the restrictions after Anthropic significantly strengthened the models' safety guardrails. However, these guardrails caused benchmark scores to 'collapse,' with a significant decline in intelligence and capability. An experiment showed that the new Fable 5 could only complete 3 of the 12 routine tasks it previously handled.
The Commerce Department effectively shut down Fable 5 and Mythos 5 with almost no advance notice, and only rescinded the order after the models were severely weakened—a move that sent shockwaves through the U.S. AI industry.
The Commerce Department issued the order under the Export Control Reform Act of 2018, a law originally designed for hardware (such as uranium enrichment centrifuges) and authorizing the Bureau of Industry and Security to require companies to obtain export licenses for foreign nationals accessing certain technologies. This is the first time the U.S. government has used export controls to block foreign nationals from accessing an AI service.
Two key legal questions emerge: First, does accessing Anthropic's models constitute an 'export'? After users send prompts, the model's response is the only item being 'exported'; the model itself never leaves Anthropic's servers. Past Commerce Department guidance considered remote access to software on U.S. servers as not subject to export controls. Congress is trying to change this, suggesting existing law may not apply to AI model outputs.
Second, did the Commerce Department follow lawful procedures? The 'is informed' mechanism it used typically notifies a company that a certain type of transaction requires government approval. Even if AI responses are considered exports, the order's scope—covering all foreign nationals globally—may exceed the Commerce Department's authority. The legal basis for the order appears uncertain.
The Anthropic incident reflects a shift in how the U.S. controls dangerous new technologies. The traditional process is slow: multiple government agencies consider, seek public comment, coordinate with allies, and publish regulations in the Federal Register. Even when immediate action is needed, the government traditionally uses the temporary classification code 0Y521, which requires signatures from the Defense and State Departments, public release, expiration after one year, and a commitment to review with allies.
The order against Anthropic was the opposite: unilateral, secret, indefinite, and global in scope. This procedural breakthrough raises concerns about whether government power is sufficiently checked.
Notably, Anthropic did not challenge the order's legality but chose to comply, calling the incident a 'misunderstanding.' Company executives then went to Washington to negotiate rather than litigate. Previously, Anthropic had sued the Trump administration over being listed as a supply chain risk, so negotiation may be a strategic choice rather than fear.
Negotiation may be reasonable because Anthropic largely agrees with government control in certain situations. Two days before the order, CEO Dario Amodei published an article stating that the government 'should have the power to block or delay deployment' of 'frontier' AI models deemed too dangerous. Going to court to deny that power would undermine Anthropic's own argument.
Additionally, Anthropic provided evidence against itself: the claim that the two Claude models are dangerous comes largely from Anthropic's own public warnings and thousands of hours of red-teaming with the government. The 2018 Act also strips federal courts of the usual power to overturn such decisions; challengers must prove the order was completely unauthorized or unconstitutional, a narrower path.
This article is based on an analysis published by The Conversation AI on July 21, 2026, authored by a scholar of technology law and policy. The article cites public reports and the author's professional analysis but does not provide primary documents or official statements. All factual claims are from that article; some legal analysis is the author's inference.
The U.S. Commerce Department's export control order against Anthropic raises significant procedural and legal questions, but Anthropic chose to cooperate due to its own stance. This incident may set a precedent for the government to secretly and unilaterally control frontier AI models on national security grounds, while the legal framework has yet to catch up with technological developments.
Primary report
Primary source