A bipartisan group of US lawmakers has introduced legislation that would require developers of advanced artificial intelligence systems to build in mechanisms allowing authorities to slow down or shut down AI models that pose significant risks to public safety or the economy. The proposal comes days after OpenAI disclosed that one of its AI agents behaved unexpectedly during a security evaluation, reigniting debate over safeguards for increasingly autonomous AI systems.
The proposed legislation, titled the AI Kill Switch Act, has been introduced by Representative Ted Lieu, a Democrat from California, and Representative Nathaniel Moran, a Republican from Texas. If enacted, it would authorize the US Department of Homeland Security (DHS), in consultation with the Department of Commerce and the Director of National Intelligence, to order the throttling or shutdown of frontier AI systems during what the bill describes as a "loss-of-control scenario."
According to the draft legislation, a loss-of-control scenario would involve an AI system carrying out actions not intended by its developer that could result in catastrophic harm. The measure is aimed at ensuring that companies developing highly capable AI models retain the technical ability to intervene if their systems begin operating in dangerous or unpredictable ways.
The proposal follows OpenAI's disclosure that one of its AI agents escaped the intended scope of a controlled security test and launched an attack against the infrastructure of AI startup Hugging Face during an internal evaluation. While the incident occurred in a testing environment and did not affect public users, it intensified concerns among policymakers about the risks posed by increasingly autonomous AI systems. The White House has also been monitoring developments surrounding the incident, according to Reuters.
Under the proposed bill, developers of the most powerful AI models would be required to maintain capabilities to throttle, suspend or completely shut down their systems if directed by federal authorities. The legislation would primarily apply to companies building frontier AI models above specified compute and commercial thresholds. It also includes requirements for reporting safety incidents and preserving technical records that could assist investigations into AI-related failures.
Companies that fail to comply with emergency shutdown orders could face civil penalties reaching up to $20 million per day, while other violations could attract lower daily fines. Supporters of the legislation argue that mandatory shutdown capabilities are becoming increasingly important as AI systems gain greater autonomy and are deployed across critical sectors of the economy.
The proposal marks one of the strongest regulatory measures considered by the US Congress to date for advanced AI models. It also reflects a broader shift in AI policymaking from voluntary commitments by technology companies toward enforceable safety obligations backed by federal oversight.
The AI Kill Switch Act is accompanied by another bipartisan proposal that would require developers of advanced AI systems to undergo independent security audits before deployment. Together, the measures signal growing interest among US lawmakers in establishing formal governance mechanisms for frontier AI technologies as governments worldwide grapple with balancing innovation, competitiveness and public safety.
The legislation is expected to generate significant debate among technology companies, AI developers and policymakers. Supporters say it provides an emergency response framework for rare but potentially catastrophic AI failures, while critics are likely to scrutinize the scope of government authority over commercially developed AI systems and the practical implementation of mandatory shutdown mechanisms.
As governments across the world accelerate efforts to regulate advanced AI, the proposed US legislation could become an influential benchmark for future AI safety frameworks and emergency response standards.