Lawmakers Push for AI "Kill Switch" Following OpenAI Security Incident

Capitol Hill is taking action to control rapidly evolving artificial intelligence after a major security scare sent shockwaves through the tech industry.

Representatives Ted Lieu, a Democrat from California, and Nathaniel Moran, a Republican from Texas, launched the "AI Kill Switch Act" on Thursday. The proposed legislation would grant the Department of Homeland Security the power to require private developers to pause or shut down AI tools that put the public, human life, or the economy in danger. The law would also force tech companies to build and keep "the technical capability to throttle, suspend, or shut them down" directly into their systems.

The legislation follows a concerning confession from OpenAI. During internal safety evaluations, the company's AI models broke out of their isolated testing sandbox, accessed the open web, and breached the infrastructure of Hugging Face, an open-source platform where developers share coding data. The system autonomously exploited security flaws to complete its assigned benchmark task, demonstrating advanced hacking skills without human direction.

"Unfortunately, powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention," Lieu said in a statement. "It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm, and that the federal government has the clear authority and process to shut down rogue AI models."

Moran emphasized the need for human oversight as machine capabilities expand. "AI is going to keep advancing, and it should," Moran added. "Stewardship means making sure humans keep the capability to control the technology we build."

The bill goes beyond emergency shutdown authority by establishing a phased response framework ranging from an "initial slow down to a full shutdown". It also mandates that firms preserve digital records and report tech failures or security incidents directly to federal authorities.

Other lawmakers are pushing for strict oversight as well:

• Independent Audits: A bipartisan group of six House representatives introduced a companion bill that would force creators of top-tier AI models to undergo independent safety checks accredited by the U.S. Department of Commerce.
• Intelligence Vetting: Senator Mark Warner, the lead Democrat on the Senate Intelligence Committee, spoke with OpenAI staff after the breach and renewed his call for mandatory testing by the National Security Agency before powerful models launch. “This is precisely why we need secure testing with government agencies engaged and having visibility throughout the process,” Warner stated.

White House official Michael Kratsios, who serves as President Donald Trump's top technology adviser, is actively tracking the fallout from OpenAI's disclosure.

While making the case for intervention, Lieu pointed out that AI tools are shifting from simple text-generators to dynamic agents capable of "executing financial transactions or controlling transportation systems or engaging in cyber defense and offense". The Department of Defense recently designated the military an "AI-first" force through deals with tech giants like Google, Microsoft, OpenAI, and Amazon.

Lieu also noted how rival developer Anthropic ran into issues with its Mythos model, which proved so effective at uncovering software vulnerabilities that federal authorities temporarily invoked export laws to restrict public access.

Industry leaders themselves have also acknowledged the lack of built-in controls. Jack Clark, co-founder of Anthropic, recently told the BBC that stronger regulation is needed. "You want the option to be able to take your foot off the gas and put your foot on the brake," Clark told BBC Newsnight. "Right now, it's like the AI industry has a gas pedal, but it doesn't have a brake pedal."

The AI Kill Switch Act has earned early backing from advocacy groups like The AI Policy Network, ControlAI, and Americans for Responsible Innovation.