When Your AI Agent Goes Wrong: The First 24 Hours of a Forensic Investigation
An autonomous agent just took an action it shouldn't have. The pager is going off, the compliance officer wants answers by Monday, and the next 24 hours will be determined entirely by the substrate that was running at the time of the action. This post is a playbook — hour by hour — for what an AI forensic investigation looks like when the primitives are in place, and what teams without them cannot answer.