Lawmakers reportedly seek records of AI models and agents gaining or attempting unauthorized access

A post quotes lawmakers criticizing delayed notification of reported breaches, saying that in several cases someone other than the developer caught the breach. The quoted letter asks OpenAI, Anthropic, Google, Meta and SpaceX to provide, by October 2, 2026, every known instance in which a model or agent they developed gained or attempted unauthorized access to a system, network, account or service outside their control.