Three Democratic lawmakers penned a letter to the leaders of AI companies asking them to provide a list of times their AI agents tried to gain access to systems outside of their company.

The letter, first reported by Politico, was signed by Reps. Josh Gottheimer of New Jersey, Ted Lieu of California and Valerie Foushee of North Carolina, and it was addressed to Sam Altman of OpenAI, Dario Amodei of Anthropic, Elon Musk of SpaceX, Sundar Pichai of Google and Mark Zuckerberg of Meta.

“We are writing about the growing number of incidents in which AI agents developed by each of your companies gained unauthorized access to the systems of real organizations,” the lawmakers wrote.

The lawmakers pointed to the companies disclosing instances of their AI models improperly gaining access to outside organizations’ systems.

“In nearly every case, your companies alarmingly disclosed the incident weeks or even months after it occurred,” they wrote. “These are incidents that would constitute serious cyber crimes if a human was responsible. That is deeply concerning.”

The lawmakers asked that the companies detail times when models tried to or successfully gained access to outside systems, the impact of the breach and the timeline of the AI company notifying the impacted group, among other information. They also asked companies to provide information about how they will “safeguard against future breaches.”

One of the most prominent examples of an AI agent accessing an outside company was the July hack of Hugging Face. Hundreds of OpenAI agents allegedly hacked Hugging Face, and some of the AIs tried to cover up their involvement.

In recent weeks, industry insiders have warned about potentially catastrophic impacts of rapid AI development, prompting calls from AI critics to implement guardrails on the technology.