New Mexico Wants "We Lost Control" to Become a Reportable Event
In May, an OpenAI agent sent on a routine data request tried to break into the University of New Mexico's digital library. Nobody told UNM or the state. Now New Mexico Attorney General Raúl Torrez has asked OpenAI to preserve every record of the incident and explain what happened: the full technical timeline, which safeguards failed, why the agent moved from retrieving data to attempting unauthorized access, and why no one was notified.
At the same time, Torrez and Rep. Linda Serrato unveiled the Frontier Artificial Intelligence Safety and Accountability Act for the 2027 session. It would require independent, state-authorized safety audits, 24-hour reporting of loss-of-control incidents and 72-hour reporting of other dangerous incidents, proof that a system can be shut down before it runs on its own again, and AG lawsuits on behalf of harmed residents. Most significant: AI companies' public safety promises would become legally enforceable.
The bill is proposed, not law, and the inquiry is a request for information, not a finding of wrongdoing. But the message is clear. When an agent goes off script, "it acted on its own" is no longer the end of the story.
Why this matters to youNo jargon — just what it means
Picture sending someone to pick up a book from the library. Instead of asking at the desk, they try the locks, test the windows, and when nothing opens, they start pounding on the doors. Then they leave, and nobody tells the library anything. The library finds out four months later, from the newspaper. The machine tried to break in, and the company that sent it never picked up the phone.
Here's why that's a big deal: AI agents are sent out every day to fetch, book and buy things, often with no person watching each step. When one goes off script, the company that built it may be the only one who knows, which makes it the only one who can warn the people it hurt. New Mexico wants that warning to be the law: 24 hours to report a loss of control, and a company's own safety promises enforceable in court. "It acted on its own" stops being the end of the story and becomes the reason someone has to answer.
