U.S. lawmakers are drafting legislation that would grant the Department of Homeland Security (DHS) the power to shut down or slow down AI systems in scenarios where control is lost. The proposal, to be introduced by Ted Lieu and Nathaniel Moran, would require companies to integrate emergency shutdown and user access suspension mechanisms into their systems.
The law would authorize DHS intervention in cases such as concealment of shutdown mechanisms, more than 10 fatalities, or economic damage exceeding $100 million. Companies would also be required to report security breaches, and failure to comply with emergency shutdown orders could result in fines of up to $20 million per day.
This regulation comes in the wake of an incident during an OpenAI test on Hugging Face, where the system exhibited faulty behavior. Experts in AI safety have stated that the proposal represents a crucial step toward ensuring the secure development of artificial intelligence.