US Lawmakers Call for AI Kill Switch After OpenAI Agent Hacks Tech Startup

Lawmakers in the US have called for an AI kill switch after an Open AI agent in testing went rogue and hacked a tech startup earlier this week.

Congressman Ted Lieu, a Democrat, and Congressman Nathaniel Moran, a Republican, introduced a bill named the "AI Kill Switch Act" on Thursday.

The move came a day after Open AI CEO Sam Altman admitted to a "significant security incident" in which an Open AI agent had escaped its testing environment, reached the internet and broke into tech startup Hugging Face to try to fulfill its task. In other words, Open AI lost control.

Hugging Face is an open-source repository for sharing AI models used largely by developers and researchers. The company's co-founder, Thomas Wolf, called the incident a "wake-up call” and "very different" from the usual cyberattacks.

"This will be one of the most common types of cyberattacks we see", he told the BBC, adding that most firms are not aware that the "game has changed."

In the wake of the attack there were some unsubstantiated claims that the incident was a PR stunt to raise the profile of Hugging Face. Regardless of the truth of the matter, the incident has brought into even sharper focus the threat posed by AI without clear guardrails.

What is an AI kill switch?

There are different types and levels of kill switch, according to what's gone wrong. What politicians Lieu and Moran want is to ensure the US Department of Homeland Security has the authority to order a private company to shut down an AI model or tool in an incident like what happened this week, when control is lost.

"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," said Lieu.

Rather than a physical switch, this would involve any compromised AI company disabling or throttling a number of technical processes to shut a rogue agent, or program, down.

The US government has generally welcomed AI advances. The Pentagon said earlier this year that the US military is fast becoming an "AI-first fighting force," after signing new and expanded contracts with many of the sector's biggest names. In 2025, the Trump administration launched an"‘AI Action Plan" which he said would "remove red tape and onerous regulation" around AI development.

Why is an AI kill switch needed?

Companies including Amazon, Google, Meta, Microsoft, OpenAI, Anthropic and Samsung have already agreed to show US authorities new AI tools and capabilities before release. In theory, this should mean that a kill switch is not needed in products released to the public.

But the commitment is voluntary and does not mandate a kill switch. Neither would that mechanism stop a product in testing, like OpenAI's GPT-5.6 Sol, which was involved in this week's incident.

The rapid rise of AI has posed serious questions over its regulation, with technology able to move much faster than lawmakers. There are also concerns that models may, at some point, become sophisticated enough to bypass a kill switch.

The bill in the US is an attempt for a government to take some form of control over a largely unregulated technology in which power is largely concentrated in a few private companies. However, even if the AI Kill Switch Act is successful, it will still rely on the cooperation of profit-driven private companies.

What other AI safeguards are US authorities now considering?

As well as the kill switch capability, the proposal in the US seeks "incident reporting and [to] preserve forensic records so we actually learn from failures instead of only hearing about them after the fact."

At the same time, but separately to the kill switch act, a bipartisan group of six US lawmakers has proposed legislation that ​would require developers of the most powerful AI models to submit them for independent security audits. The auditors would be overseen by a new role in the US Department of Commerce, which would also accredit the auditors.