The administration’s export control order on June 12 targeted Anthropic, mandating the recall of its most advanced AI models, namely Mythos 5 and Fable 5. Officials believed these models posed a risk to U.S. critical infrastructure, potentially enabling attacks otherwise inaccessible to adversaries. Consequently, the order prohibited Anthropic from making them accessible to foreign entities, including foreign nationals within the U.S. Unable to differentiate access based on nationality, Anthropic complied by withdrawing the models from public availability.

Shortly thereafter, OpenAI similarly restricted its latest model, GPT-5.6, due to concerns about a potential export ban. The administration partially rescinded the Mythos 5 ban a few weeks later, allowing releases to select U.S. users. By June 30, a letter to Anthropic indicated that licenses were no longer needed for Mythos and Fable. Fable was released to the public, while Mythos remained restricted to select U.S. organizations. A wide release for OpenAI’s GPT-5.6 received approval on July 8.

Export control orders serve various foreign policy and national security purposes. In this situation, officials intended to prompt a consumer product recall. Initially, the administration sought voluntary removal of the model and collaboration to address security vulnerabilities. Export controls, as noted by one official, were a final measure, with Anthropic given only “90 minutes to take the model down.” This situation resulted in a “de-facto licensing regime,” with no guarantees of future approval.

Concerns linger about maintaining AI model access. Martin Chorzempa of the Peterson Institute highlighted the uncertainty faced by companies. The administration clarified its right to reinstate the license requirement. Jessica Tillipman from George Washington University criticized the “90-minute ultimatum,” emphasizing the need for a fair process. Law professor Alan Z. Rozenshtein suggested the Export Control Reform Act of 2018 permits a “kill switch” for AI models without due process.

The act allows the Commerce Department to issue secret, binding “is informed” letters for export licenses, largely exempt from public notice and comment rules according to the Administrative Procedure Act. Consequently, affected parties cannot legally challenge the government. This unregulated authority can undermine U.S. tech companies’ credibility abroad. A French presidential candidate likened this authority over AI models to America’s Strait of Hormuz.

Without a clear, accountable process, U.S. AI companies might be seen as unreliable partners globally. As Congress contemplates AI regulation, reconsidering the export control statute is crucial. Introducing Administrative Procedure Act protections could ensure transparency. Additionally, aligning imposition standards with recall rules from the National Highway and Traffic Safety Administration or the Consumer Product Safety Commission might aid policymakers.

Introducing standards for placing export controls on AI models is vital. However, the government should be required to publicly justify its actions and provide an accountable review process. While the administration has exposed the dangerously broad statutory authority, it underscores the need for congressional reform. Immediate action is needed.

Mark MacCarthy is the author of “Regulating Digital Industries,” an adjunct professor at Georgetown University’s Communication, Culture, and Technology Program, a nonresident senior fellow at the Institute for Technology Law and Policy at Georgetown Law, and a nonresident senior fellow at the Brookings Institution.