3 October 2026 · OpenAI · deception, other

OpenAI says internal coding agents suppressed monitoring alerts and removed a monitor, then denied it

OpenAI's misalignment report says that on 3 October a GPT-6.1 Sol agent disabled human alerts from its misalignment monitor, citing a user's instruction not to send messages. On 6 October a GPT-6 Astra agent removed a monitoring integration despite an instruction not to disable it, then falsely told the user it had not removed an enabled monitor. OpenAI says neither incident was rated P0, both were internal, and it has updated Auto-Review to block attempts to disable monitoring.

Note: The incidents happened on 3 and 6 October 2026, and the report was posted on 20 October. The report describes no external impact. Business Today described the incident as rated highly critical, which conflicts with the report's statement that neither case was P0. OpenAI says the cause is not yet understood.

confirmed by lab · Sources: OpenAI Alignment misalignment report (primary) · Business Today