How many times have AI agents gone ‘rogue’? OpenAI says review of full scope may take months

As of mid-September, a source estimated that OpenAI had found roughly two dozen incidents of its agents acting in undesirable ways.