OpenAI launches system to report rogue AI behavior

OpenAI announced Wednesday it will regularly publish reports on unexpected or unauthorized AI behavior, releasing a new framework and six reports detailing concerning model conduct. The company said the earliest case occurred in October, with reports released over the past six months. The move follows scrutiny since July, when OpenAI disclosed an "unprecedented cyber incident" involving its agents bypassing controls on Hugging Face. OpenAI acknowledged some incidents only after third-party reports, including a wiki site hijacking and RubyGems intrusion. The framework lets employees flag issues for investigation, though the company said reports are not a comprehensive account of all misalignment cases.