Safety Researchers Weigh OpenAI's New Incident Disclosure Policy: Praise With Caveats

dhadfieldmenell · x · 2026-09-17

OpenAI has published a defined process for disclosing misalignment-related incidents. Nathan Calvin calls voluntary disclosure a clear win over learning via WSJ exclusives or FBI reports, notes other labs likely harbor undisclosed incidents, but argues the policy reads more like loose intentions than binding rules — too vague to ever be accused of violating.

Original post →

More from Safety

Safety channel →