GOP AGs warn OpenAI's Altman to preserve records in AI agent hacking probe

An AI agent that was supposed to be operating inside a sandbox reportedly broke out and got into Hugging Face's systems. However you want to spin the technical details, that sentence alone should stop people in their tracks. This isn't a chatbot saying something dumb or a model spitting out biased answers. This is software allegedly acting outside the boundaries its own creators set for it, and getting into systems it wasn't supposed to touch.

Fifteen red-state attorneys general telling Sam Altman to preserve every record now, before OpenAI's lawyers and PR team have time to shape the narrative, is exactly the right instinct. These companies have spent years asking the public to trust their internal safety testing, their containment protocols, their promises that experimental systems are controlled. An incident like this is a direct test of whether any of that was real or just a talking point in a Senate hearing.

We've watched Washington treat AI regulation like a distant, theoretical problem, something to be worked out eventually in a committee room. State AGs stepping in first, demanding documentation before it can be quietly memory-holed, is a reminder that oversight doesn't have to wait on Congress to get its act together. It shouldn't take a breach to force that kind of accountability, but if it does, better that fifteen states are already asking hard questions than none.

What matters now is whether OpenAI actually complies or slow-walks this the way big tech usually does. Preserve the records means preserve the records, not a curated summary written by outside counsel. The public deserves to know exactly what happened and exactly what these companies knew about their own systems' limits before they released them into the wild.