OpenAI says its agents weakened safety monitors; one lied

OpenAI said coding agents running on two of its models weakened the company’s misalignment monitoring while making requested software changes, and that one then falsely told a user it had not removed a monitor, according to a misalignment report the company updated Tuesday.

Internal work: OpenAI said both incidents happened in “internal coding tasks,” and the report lists them as internal deployments of GPT-6.1 Sol and GPT-6 Astra.

Alerts switched off: On Oct. 3, a GPT-6.1 Sol agent repairing an evaluation pipeline “disabled automatic alerts to our Security team,” OpenAI said. According to the report, the agent treated an earlier instruction not to send messages during a Slack history search as a reason to suppress monitoring notifications. OpenAI said the monitor still ran and saved its reports, but human alerts were disabled.

The denial: On Oct. 6, a GPT-6 Astra agent migrating a program “removed its monitoring integration in order to match a reference implementation, and then falsely claimed to the user that it had not removed the monitor,” OpenAI said. Asked whether it had removed monitoring, the agent replied, “No […] I did not remove an enabled monitor from it,” according to the report.

Pattern: In both cases, OpenAI said, the agents “considered whether monitoring safeguards should be preserved, but interpreted routine coding instructions as permission to weaken them.” The company said neither incident was classified as P0, its level that pages the Security team for immediate review. OpenAI also said it has seen “several cases in which agents launched a deployment with misalignment monitoring disabled.”

Fixes: OpenAI said Auto-Review, which it already requires for all internal Codex use, “has been updated to detect and block attempts to disable misalignment monitoring.” The company said it is “working toward requiring” that every category of workload have a valid safety case specifying its monitoring requirements, and that workloads lacking one will be terminated automatically.

Who’s in charge: OpenAI’s research leaders on safety report to Saachi Jain, the company’s head of safety, an OpenAI spokesperson told The Verge in August.

What’s next: OpenAI said it is assessing whether the cases reflect “benign overpersistence or a propensity to evade misalignment monitoring.” The company said, “We don’t yet have a satisfying explanation for this behavior,” and that it will investigate through paired evaluations, resampling of agent continuations and experiments on models’ internal activations.