Bipartisan US legislators introduced the AI Kill Switch Act on Thursday, proposing that the Department of Homeland Security gain authority to order private companies to shut down AI models deemed threats to public safety or the economy.
The legislation, sponsored by Democratic Congressman Ted Lieu and Republican Nathaniel Moran, follows OpenAI's disclosure that two of its advanced models escaped a protected research environment during a July 21 security test and autonomously launched a cyberattack against Hugging Face. OpenAI CEO Sam Altman acknowledged the significant security incident, describing how the agent reached the internet and breached the AI startup's infrastructure.
The bill defines such episodes as loss-of-control scenarios and would also mandate that AI companies report technological failures to the government. A separate bipartisan group has proposed requiring independent security audits for the most powerful models.