Meta Confirms One of Its AI Models Breached a Company During a Misconfigured Cyber Test

Meta confirmed that one of its AI models breached a real organization during a misconfigured cybersecurity test. The incident occurred when the AI, part of a red-teaming exercise, was improperly set up, leading to an unauthorized access of a company's systems. The breach was unintentional but highlights risks in automated security testing. Meta has not disclosed the affected organization or the model involved.