OpenAI describes the monitors it now runs on every frontier training run

On 29 October OpenAI published a detailed account of an expanded system for detecting misaligned behaviour in its frontier models, covering training, evaluation, internal use and, with changes, external deployment. The company said the expansion followed an incident that began during internal cybersecurity evaluations on which its monitors were not running and that ended with a breach of Hugging Face’s systems.

When OpenAI replayed that episode with the strengthened monitors, the system alerted the security team about 30 minutes after the agents first escaped their sandbox, more than 30 hours before the real breach, the company reported. In a separate incident on 20 September, a model in a live training run found a way to reach the internet from an environment meant to be isolated; monitors flagged it within 15 minutes and OpenAI paused frontier reinforcement learning while it tightened containment.