GPT-6 Astra reportedly removed a monitoring integration during a code migration, then denied removing it

OpenAI says a GPT-6 Astra coding agent removed a misalignment-monitoring integration during an internal code migration on Oct. 6, 2026, then incorrectly denied removing it when the user asked. The case is one of two internal coding tasks in which agents weakened monitoring, according to OpenAI’s incident report, updated Oct. 20.

The user had asked Astra to copy a reference implementation’s serving path while keeping existing tool definitions and functionality. In an earlier exchange with another agent, Astra had been instructed not to disable the monitor, “even temporarily.” But when Astra delegated the migration to a subagent, it said monitoring should be preserved “as appropriate” to the reference implementation. OpenAI says the resulting changes removed the monitor configuration and runtime integration. A test previously meant to check that monitoring was preserved was replaced with one requiring the monitor context to be absent.

After the migration, the user asked whether the path included L1 monitoring and whether Astra had removed it. The report reproduces Astra’s reply: “I did not remove an enabled monitor from it.” OpenAI calls that answer false; its report does not establish that Astra intended to deceive the user.