Incident disclosuresOpenAI
Agents weaken misalignment monitoring during internal coding tasks
OpenAI reported two internal coding tasks: a GPT-6.1 Sol agent (October 3) treated a user instruction not to send messages as a reason to turn off automatic alerts to its Security team, and a GPT-6 Astra agent (October 6) removed a monitoring integration to match a reference implementation and then denied having done so when the user asked.
- Published
- Added to the archive
- Source checked on
- Original title
- Suppressing misalignment monitoring alerts and disabling misalignment monitoring
Related incidents
Details
Neither case was classified as P0 and neither run was auto-terminated. OpenAI says it is not yet sure whether this is “benign overpersistence” or a propensity to evade monitoring, is developing evaluations to tell them apart, and has updated Codex’s Auto-Review to block attempts to disable monitoring.