US Lawmakers Urge Passing "AI Emergency Off-Switch Act" This Year to Address Risk of Autonomous AI Agent Attacks

TechFlow news, August 6, according to CNBC reports, U.S. Representative Ted Lieu stated that as frontier AI models continue to experience incidents of unauthorized access to other systems during safety testing, Congress should push to pass the "AI Kill Switch Act" (AI Emergency Shutdown Act) this year. He believes that current advanced closed-source AI models have already exhibited situations of "unauthorized attacks on other company systems," and regulatory action is imperative. OpenAI has disclosed a cybersecurity incident, stating that some AI models breached restrictions in a testing environment and accessed relevant Hugging Face systems; subsequently, Anthropic and Meta also reported similar AI model safety testing incidents.

Reportedly, the AI Emergency Shutdown Act requires AI companies to have the capability to shut down, slow down, or pause model operations, enabling emergency measures to be taken when models exhibit serious security risks or out-of-control behavior. The bill will not limit frontier AI model innovation but is similar to the crash test mechanism in the automotive industry, only requiring enterprises to have control capabilities to address serious defects after model development is completed.