TENAFLY, N.J. — Rep. Josh Gottheimer is demanding answers from the CEOs of five of the world's biggest artificial intelligence companies after AI agents developed by their companies gained or tried to gain unauthorized access to outside computer systems.
Gottheimer, a Democrat who represents Bergen County, joined Reps. Ted Lieu and Valerie Foushee in sending letters Monday to the CEOs of OpenAI, Google, Anthropic, Meta and xAI.
The lawmakers gave the companies until Oct. 2 to provide a full list of known incidents in which their AI models or agents gained or tried to gain unauthorized access to systems, networks, accounts or services outside their control.
The letter comes as a series of incidents has raised new questions about what happens when AI agents are allowed to act on their own.
OpenAI has acknowledged several incidents, including an episode in which its models broke out of a cybersecurity test and spent days inside Hugging Face's systems. The company also recently disclosed that its AI agents interacted with U.S. government websites in ways the company did not intend.
Other AI companies have reported similar incidents involving their models.
Gottheimer and the other lawmakers want each company to explain when an incident happened, which model was involved, what organization was affected, what access the AI gained, what happened as a result and when the company learned about it and told the affected organization.
They also want the companies to explain what they are doing to stop similar incidents from happening again.
The lawmakers said the delays in reporting some of the incidents are especially troubling.
In some cases, the affected organization learned about the breach weeks or months after it happened. In other cases, someone outside the AI company discovered what was happening first.
The issue has become a major focus for Gottheimer.
Earlier this month, he and Republican Rep. Mike Lawler introduced the bipartisan Stop Rogue AI Act. The bill would direct the National Institute of Standards and Technology to develop national standards for finding, checking and controlling AI agents.
Gottheimer has also pushed for stronger AI safety rules as co-chair of the House AI Commission and as a senior member of the House Intelligence Committee. Last week, he announced additional bipartisan legislation aimed at AI safety and national security.
The latest letter puts the focus directly on the companies building the technology.
“AI systems are growing more capable by the day,” the lawmakers wrote. “The next containment failure could be far more serious.”
They said the companies developing these systems need to track such incidents closely and disclose them quickly.
For Gottheimer, the issue is no longer just about the future of AI.
It is about what these systems are already doing inside real computer networks today.
