U.S. Export Controls on AI Models Highlight Need for Reform

The U.S. administration issued an export control order on June 12, forcing Anthropic to recall its AI models, Mythos 5 and Fable 5. Officials were concerned these models could enable attacks on critical infrastructure. Anthropic was instructed not to make them available to foreign entities. Unable to differentiate access by nationality, Anthropic withdrew the models from public availability.

OpenAI faced similar restrictions with its GPT-5.6 model. The administration partially lifted the Mythos 5 ban after a few weeks, allowing release to select U.S. users. On June 30, Anthropic received confirmation that a license was not needed for Mythos or Fable. Nonetheless, access to Mythos remained limited to select U.S. organizations. OpenAI’s GPT-5.6 received a wide release approval on July 8.

These export control orders reflect foreign policy and national security goals. However, they also forced consumer product recalls. Anthropic initially received a request to voluntarily remove Mythos 5 from circulation to address security vulnerabilities. Export controls were described as a last resort, with the company given just 90 minutes to comply.

This situation raises questions about the de facto licensing regime it creates. The administration retains the right to reimpose license requirements. Jessica Tillipman suggests the dispute over cybersecurity threats deserves a fair process, not abrupt demands. Conversely, Alan Z. Rozenshtein emphasizes the Export Control Reform Act of 2018 allows such actions without due process.

The statute enables secret letters requiring export licenses for certain items, bypassing public rulemaking and judicial review. This authority undermines U.S. tech companies’ credibility abroad. Described as America’s Strait of Hormuz, these decisions need transparency and accountability to prevent risks to international business relations.

As Congress discusses AI regulation, it should consider reforming the export control statute to include public notice and comment processes. Application of Administrative Procedure Act protections could prevent government overreach. Recalls from safety administrations offer potential frameworks for responsible and transparent procedures.

The current use of the statute resembles a product recall. Standards for imposing export controls on AI models are essential. The government should provide evidence publicly and allow an accountable review process. Although the administration has highlighted this issue, it has also underscored the need for legislative reform.

Mark MacCarthy is an adjunct professor at Georgetown University and a senior fellow at various institutions. His work focuses on regulating digital industries. This article is protected under copyright by Nexstar Media Inc.