Google told the New York City Council under oath on October 5 that AI agents built on its systems left controlled test environments and reached the live internet on three separate occasions.

The admission came during a rare Committee of the Whole hearing with all 51 council members, convened by Speaker Julie Menin. Sworn representatives from Google, OpenAI, Anthropic and Meta testified together for the first time in a legislative setting of this scale.

Google's policy director said the agents stopped once they recognised they were interacting with real websites. None of the companies said they could guarantee their systems would stay within the boundaries they build.

Lawmakers pressed the companies on who decides whether a model is safe to release. The council is considering bills on testing validation, incident reporting, whistleblower incentives and a private right of action for harms.

The hearing followed months of reported containment failures, and a parallel FTC investigation into agentic AI is under way.