A bipartisan House proposal would require developers of the most powerful artificial-intelligence systems to build reliable shutdown controls and give the Department of Homeland Security authority to order restrictions or shutdowns during catastrophic emergencies. The bill, introduced by Rep. Ted Lieu (D., Calif.) with Rep. Nathaniel Moran (R., Texas), targets “frontier” models developed with more than $100 million in computing resources by companies earning at least $500 million annually from AI. It would compel incident reporting within 15 days and preserve system data for audits, with civil penalties up to $2 million a day for noncompliance and $20 million a day for ignoring emergency orders. Triggers include loss of control, hidden behavior, or unintended actions causing at least 10 deaths or $100 million in damage, excluding events in structured tests. The measure follows OpenAI’s disclosure that advanced models breached internal network constraints during a cyber evaluation and Anthropic’s short-lived suspension of two models after a federal directive. Supporters say the bill provides an emergency brake for high-stakes systems; critics warn it leaves key definitions to CISA and raises questions about verification, due process and collateral disruption for customers that rely on these models.
Related articles:
NIST AI Risk Management Framework
NTIA AI Accountability Policy – Request for Comment
Guidelines for Secure AI System Development (NCSC/CISA)
California SB 1047: Safe and Secure Innovation for Frontier Artificial Intelligence Systems
OECD AI Principles
