OpenAI tells lawmakers it is building automated shutdown capabilities

A letter to US lawmakers says OpenAI is developing automated shutdown capabilities, stronger monitoring and tighter internet restrictions for safety tests after an agent-related incident. Lawmakers continued to seek fuller logs and accountability.

Why it matters: Agent control is moving beyond prompt-level safeguards towards runtime containment and interruption.

Business/technical/strategic impact: Enterprises need independent control planes that can revoke tools and credentials even if a model or vendor service misbehaves. The design, coverage and test evidence for the shutdown capability were not public.