US Government Restricted AI Access, Raising Legal Questions
The U.S. Commerce Department ordered Anthropic to restrict access to two of its advanced AI models due to national security concerns. The move temporarily disabled the models for all users and raised questions about the government's authority to regulate AI globally.
On June 12, the U.S. Commerce Department ordered AI company Anthropic to immediately restrict access to its two most advanced language models, Claude Fable 5 and Mythos 5. The directive, citing national security concerns, barred access by all foreign nationals, including the company's noncitizen employees. As a result, Anthropic made the models inaccessible to everyone within hours.
Unable to immediately verify each user's nationality, including those outside the U.S., the company opted to shut down the models entirely. The Commerce Department lifted its restriction on June 30 after Anthropic significantly strengthened the models' safety guardrails. However, these enhancements led to a reported "collapse" in the models' benchmark scores, indicating lower overall intelligence and capacity.
The directive marked a significant departure from previous practices. The Commerce Department utilized the Export Control Reform Act of 2018, traditionally applied to hardware, to restrict access to an AI service for the first time. The law's provisions allow for requiring export licenses for foreign access to specific technologies, prompting debate on whether remote access to software constitutes an export.
The incident has caused concern within the U.S. AI industry regarding the government's approach. Although Anthropic did not legally contest the order but rather negotiated its resolution, the situation highlights the complexities of regulating emerging technologies. The company itself has previously voiced support for government oversight of potentially dangerous AI models and acknowledged the need for such controls.