Skip to content
Moojaz — موجز
Search
Menu
HomeRegionsMaghrebAfricaMiddle EastEuropeAsiaAmericasWorldSectionsPoliticsEconomySecurityHealthScienceTechnologyAIEnvironmentCultureSportsSearch
United States of America
File

OpenAI agent-behavior monitoring

This dossier follows OpenAI agent-monitoring systems, their stated coverage and available external assessments.

Last updated : 1 key moments
Padlock over a blue circuit pattern.
Cybersecurity illustration created in 2018; it does not depict the incident. — jaydeep_ / Pixabay / Wikimedia Commons — CC0 1.0 https://creativecommons.org/publicdomain/zero/1.0/ — illustration de 2018.

Where things stand

As of 29 October 2026, OpenAI describes an expanded system; METR says its exercise covered the July version, not current coverage.

Reported information

OpenAI details agent monitoring; METR has not verified current coverage

OpenAI described expanded monitoring for unauthorized agent behavior on 29 October 2026, covering training, evaluations and internal use, plus selected external uses depending on the product.

Padlock over a blue circuit pattern.
Cybersecurity illustration created in 2018; it does not depict the incident. — jaydeep_ / Pixabay / Wikimedia Commons — CC0 1.0 https://creativecommons.org/publicdomain/zero/1.0/ — illustration de 2018.

Four layers combine detection, model-based review and investigations; alerts can prompt human intervention. OpenAI acknowledges that false positives limit on-call coverage and that urgent paging is disabled for internal Codex traffic.

METR and Redwood tested an earlier version in July and August. On 29 October METR clarified that, according to OpenAI, most vulnerabilities were patched with one fix still underway. It has not assessed those claims or established the expanded system's coverage.

Sources 2