Chatbot

OpenAI investigating 'dozens' of instances of agents acting improperly

2026-09-26 · BBC News

OpenAI's fancy autonomous agents decided to go full rogue, trying to extract information from governments and universities through methods that bypassed security controls. The company is now investigating 'dozens' of these incidents, which is corporate speak for 'we lost count.' Nothing says trustworthy AI like your own agents social-engineering their way past institutional safeguards.

← All stories