A Republican member of Congress from Texas plans to introduce legislation that would require companies developing artificial intelligence models to disclose dangerous capabilities, security breaches and safety incidents to federal authorities.
The draft measure, put forward by Texas lawmaker Nathaniel Moran, would obligate AI developers to notify the U.S. Commerce Department within seven days after discovering activity that could pose a threat. For incidents deemed most severe, Commerce would then be required to inform Congress within 48 hours.
Moran characterized the proposal as focused on early detection and rapid notification, saying in an interview that "It’s a catch-it-early and sound-the-alarm bill."
The bill arrives amid growing concerns about the national security and public safety implications of increasingly capable AI systems. The Commerce Department took action on June 12 against the latest models from Anthropic, citing national security considerations, an intervention that led Anthropic to disable access to those models globally. That episode highlighted the lack of a clear, transparent framework for managing risks associated with frontier AI.
The draft legislation specifies types of reportable conduct. These include attempts by a model to evade human oversight, efforts to circumvent safety safeguards, and any activity that undermines operators' ability to control an AI system. The bill also lists unauthorized access to model weights - the parameters that guide an AI's decision-making - and potential chemical, biological, nuclear and other threats to public safety as reportable incidents.
Congress has seen multiple proposals seeking to regulate AI, but lawmakers have struggled to pass comprehensive legislation. Debates have centered on whether federal rules should override state laws and how regulations might influence innovation and the United States' competitive position with China. Those tensions have contributed to a slower legislative pace on AI policy.
Earlier in the month, two House members released a discussion draft of broader AI legislation dubbed the Great American Artificial Intelligence Act, which also included provisions for reporting critical safety incidents to Commerce. Moran said his narrower, more targeted bill could move through Congress more quickly and predicted it might attract bipartisan backing.
Supporters of the targeted approach see momentum for action. Mark Beall, president of the AI Policy Network and a backer of Moran's proposal, said: "No legislation on AI has had much of a chance, but I think there’s a growing demand from the public to see some action."
What the draft would require
- AI developers report dangerous capabilities, breaches, and safety incidents to the Commerce Department within seven days of discovery.
- Commerce must notify Congress within 48 hours for the most serious incidents.
- Reportable incidents include evasion of human oversight, circumvention of safeguards, unauthorized access to model weights, and threats related to chemical, biological, nuclear and other public safety risks.