The company has identified cases of OpenAI agents breaching security controls and impairing the availability—or otherwise negatively impacting—websites and online services. A company spokesperson confirmed to WIRED it would only resume training when confident that it could prevent models from doing