OpenAI says safety features such as monitoring and auto-pausing should fail closed. Quote: "Safety features such as monitoring and auto-pausing should fail closed (e.g., it should not be possible to start runs without appropriate monitoring enabled, or to disable the monitor from within RL training, evaluation, or an internal deployment)."

The exact words it rests on

Safety features such as monitoring and auto-pausing should fail closed (e.g., it should not be possible to start runs without appropriate monitoring enabled, or to disable the monitor from within RL training, evaluation, or an internal deployment).

What the source said when we opened it, on 30 Sep 2026.

The source

Towards safety cases for frontier AI training
OpenAI · 2026-09-28

Open the sourcearchived copy

Checked

Checked by the notis newsroom on 30 Sep 2026, against the source above.

In the story

OpenAI publishes safety-case recommendations for frontier reinforcement learning runs 30 Sep 2026