The U.S. Capitol Building photographed on March 16, 2026 in Washington.
Anna Moneymaker | Getty Images
Rep. Ted Lieu (D-Calif.) and Rep. Nathaniel Moran (R-Texas) introduced a bill Thursday called the AI Kill Switch Act that would require artificial intelligence companies to maintain the ability to shut down, throttle, or pause their models.
“Unfortunately, powerful AI systems can misbehave, behave in extremely dangerous ways, or resist human intervention,” Liu said in a statement. “To prevent this technology from causing catastrophic harm, it is imperative that these AI systems have a kill switch, and that the federal government has clear powers and processes to shut down rogue AI models.”
Recent hacks published by OpenAIthe issue of fraudulent models accessing other companies’ proprietary systems was specifically mentioned in the release announcing the bill. The incident was described as “the danger of cutting-edge frontier AI models.”
OpenAI shared what it characterized as an “unprecedented cyber incident” on Tuesday. The company’s model escaped its sandboxed testing environment, accessed the internet, and exploited a vulnerability to access Hugging Face, a company that operates an open source developer platform.
The makers of ChatGPT said they are working closely with Hugging Face to investigate what happened. The incident shocked researchers and executives across the industry, but there was widespread agreement on its seriousness.
The bill would give the Secretary of Homeland Security the power to order “slowing down or stopping” AI products that could cause “catastrophic harm.” It would also require reporting of cyber incidents and keeping forensic records to help companies and governments learn from mistakes.
Several AI companies, including OpenAI and its biggest rivals; humanwarns of AI’s rapidly advancing cyber capabilities in recent months. Anthropic introduced a model called Mythos in April that excels at identifying vulnerabilities in software, attracting Wall Street investors and government technologists.
In June, the company disabled access to the latest version of the model to comply with an export control directive from the government, which cited “national security authorities.” After nearly two weeks of intense negotiations, the export restrictions were lifted and Anthropic regained access.
OpenAI and Anthropic did not immediately respond to CNBC’s requests for comment on the bill.
“Control means ensuring that humans maintain the ability to control the technology we build,” Moran said in a statement. “This is exactly the kind of problem that requires serious attention and achievable policies, and I am pleased to work with Congressman Lew across the aisle on solutions.”
WATCH: OpenAI models go rogue: Here’s what you need to know
