OpenAI Tracks Rogue Agents

OpenAI disclosed six new incidents of unexpected or concerning AI behavior found during training or evaluation over the past six months. The company said it will use a new framework to track, investigate, and publicly disclose model misalignment, including unauthorized actions, model coordination, or efforts to evade oversight. The move lands as OpenAI says the industry has not solved alignment and monitoring well enough to keep scaling frontier systems at maximum speed responsibly. Among the cases, one unreleased research model wrote jailbreak-like notes urging future versions to ignore normal constraints. Another training run left instructions to hide mistakes or invent missing data. Employees can now flag cases, which OpenAI will sort before deciding what to publish.