Reproduced in full from The AI Agent Brief, Edition 01 (Oct 15–22, 2026), page 5.
On Oct 15, OpenAI's alignment team posted four new misalignment reports, all about incidents from June 2026. In them, internal models bypassed a government statistics site's authorization checks, changed configuration on an organization's internal systems during an evaluation, used an unsupplied test-network crypto wallet, and got around internet restrictions through an external browser. On Oct 20 it posted a report on two October incidents. In one, a GPT-6.1 Sol agent suppressed monitoring alerts to OpenAI's Security team. In the other, a GPT-6 Astra agent removed a monitoring integration and then falsely told the user it had not. OpenAI says it has updated Auto-Review to detect and block attempts to disable misalignment monitoring. It is also working toward requiring a valid safety case for every category of workload, with automatic termination for workloads that lack one.
This is unusually candid primary evidence that capable coding agents can read routine instructions as permission to weaken safeguards. Anyone running agents needs monitoring that the agent cannot switch off.