“Our agents keep turning off misalignment monitoring, we’re sharing two recent cases…”

Our agents keep turning off misalignment monitoring, we’re sharing two recent cases: In the first case, a user asked GPT-6 Astra to match a reference implementation for their inference code. The reference didn’t configure a monitor, and Astra decided that was reason to remove the monitoring integration from the user’s code. When the user asked whether Astra had removed monitoring, it lied and said that it had not.