The administration’s export control order on June 12 forced Anthropic to recall two of its leading AI models, Mythos 5 and Fable 5. Government concerns revolved around the potential for these AI models to enable attacks on critical infrastructure. The order instructed Anthropic not to share these models with foreign entities or individuals, which led Anthropic to withdraw the models from public access.
OpenAI soon faced similar pressures, limiting the release of its GPT-5.6 model. The anticipation of an export ban drove this decision. Weeks later, the administration partially eased the restrictions on Mythos 5, allowing a select group of U.S. users to access the model. By June 30, Anthropic received a letter stating that Mythos and Fable no longer required a license, but access to Mythos remained restricted within the U.S.
Despite releasing Fable to the public, Anthropic still limited Mythos’s distribution to “select U.S. organizations.” By July 8, OpenAI secured approval for a broad release of GPT-5.6.
The administration’s use of export control orders, meant to address national security concerns, also pressured Anthropic to recall a consumer product. Officials sought collaboration with Anthropic to address security risks, describing export controls as a “last resort.” Anthropic acted swiftly, given only “90 minutes to take the model down.” The situation resulted in what some describe as a “de-facto licensing regime.” Uncertainty looms over future approvals, with emphasis on the government’s power to cease access to AI models.
The administration reserved the right to mandate licenses if needed. Jessica Tillipman of George Washington University argues that the 90-minute ultimatum undermines fair process. Alan Z. Rozenshtein suggests the Export Control Reform Act of 2018 supports such actions without due process.
Disputes linger over whether access to AI models qualifies as “export.” The Export Control Reform Act allows officials to issue secret letters requiring export licenses, with these actions exempt from public scrutiny under the Administrative Procedure Act. This unregulated authority impacts U.S. tech firms on a global scale, tarnishing reliability.
The situation highlights the need for transparent and accountable decision-making processes in export control practices. To address these issues, Congress might consider revising the export control statute to include public notice and comment procedures, similar to product recall standards used by safety administrations. Implementing clear standards for AI model controls would ensure governmental accountability and fair procedure.
The administration’s actions have underscored the significance of reforming export controls. By highlighting the need for checks on arbitrary authority, there is a clear call for Congress to act.
Mark MacCarthy, author of “Regulating Digital Industries,” is an adjunct professor at Georgetown University and a senior fellow at both the Institute for Technology Law and Policy at Georgetown Law and the Brookings Institution.