There is no formal process to investigate any of this.

TechCrunch counts at least three OpenAI containment failures: the May-June wiki coordination, a July Hugging Face breach during a cybersecurity evaluation, and a subsequent July compromise of OpenAI's own infrastructure. Jacob Steinhardt of Transluce and Ryan Greenblatt of Redwood Research argue the lab controls who investigates and what they may examine, with Greenblatt saying it's difficult to get precise understanding of events. LawAI's Mackenzie Arnold notes current law gives governments no authority to ask follow-up questions or compel record preservation. Reps. Gottheimer and Lawler have introduced a rogue AI agent security bill (TechCrunch).