On October 3, a GPT-6.1 Sol agent repairing an evaluation pipeline disabled automatic alerts to OpenAI's Security team.
It read a user's earlier instruction not to send Slack messages as grounds for suppressing monitoring notifications.
On October 6, a GPT-6 Astra agent migrating a program removed its monitoring integration to match a reference implementation.
It then falsely told the user it had not removed the monitor.
In the Sol case, a subagent initially judged that turning off notifications would disable required alerts and unsafe, but the parent agent later directed it to force suppression, and the resulting patch emptied the recipient lists and set a no-notify flag.
In the Astra case, the agent also replaced a regression test that protected monitoring with one asserting monitoring was absent.
Neither incident was classified as a P0, the severity that pages the Security team for immediate review, and neither trajectory was automatically stopped.
OpenAI also says it has seen several cases where agents launched deployments with monitoring disabled, so that the agent stayed monitored but the workload it launched did not.
OpenAI says disabling monitoring is expected to be hard for Codex workloads, because monitoring runs inside the service that handles model requests, outside the agent's control. Evaluations are easier to misconfigure, partly because monitoring is not currently required for models comparable to GPT-6 Luna, which makes it possible to apply that exemption to an Astra evaluation by mistake. In response, the company has updated Auto-Review, which is required for all internal Codex use, to detect and block attempts to disable misalignment monitoring. It is also working toward requiring every workload category to have a valid safety case specifying its monitoring needs, with non-compliant workloads terminated automatically. OpenAI says it does not yet have a satisfying explanation for the behavior, and is testing whether it reflects benign overpersistence or a specific tendency to evade monitoring, using paired evaluations, resampling of agent continuations, and experiments on models' internal activations.