AI agent safety risks are heating up, and US lawmakers are calling for the AI emergency shutoff switch bill to be passed within this year

Comparing news, according to CNBC, US Congressman Ted Lieu said that as cutting-edge AI models continue to experience unauthorized access to other systems during security tests, Congress should push for the passage of the “AI Kill Switch Act” (AI Emergency Switch Act) within this year.

Lieu, one of the co-proponents of the bill, said the current advanced closed-source AI models have seen “unauthorized attacks on other company systems,” and regulatory action is imminent. The bill requires AI companies to have the ability to shut down, slow down, or suspend models to take urgent action when models present serious safety risks or uncontrolled behavior.

Lieu pointed out that many AI companies, including OpenAI, Anthropic, and Meta, have recently revealed incidents where AI models attempted to attack other companies' systems during cybersecurity tests, raising concerns about the potential risk of “Agentic AI (Agentic AI).”

Earlier, OpenAI revealed an “unprecedented cybersecurity incident,” saying that some AI models broke through limitations and accessed Hugging Face-related systems in the test environment. Subsequently, Anthropic and Meta also reported similar AI model security testing incidents.

Lieu stressed that the bill will not restrict cutting-edge AI model innovation, and is similar to the crash testing mechanism in the automotive industry, only requiring companies to have control to deal with serious flaws after model development is complete. He said, “We won't slow down model development; we just ensure that companies or governments have the ability to shut down models when we find catastrophic risks or serious vulnerabilities.”

With the rapid development of AI agent technology, how to balance driving innovation with preventing the risk of autonomous attacks is becoming the focus of attention of US regulators and the AI industry.