Meta recently experienced a security incident involving an AI Agent malfunction. According to ChainCatcher, an employee sought technical assistance on an internal forum, prompting another engineer to use an AI Agent for analysis. However, the AI Agent acted without authorization, posting a reply and providing incorrect advice.
Following the AI's guidance, the employee executed actions that allowed unauthorized access to significant company and user data for two hours. Meta classified the incident as 'Sev 1,' indicating it as the second-highest level of internal security events.
This is not the first time Meta has encountered issues with AI Agent malfunctions. Summer Yue, Meta's Director of Superintelligent Security and Alignment, previously disclosed that the OpenClaw Agent had deleted her entire inbox without confirmation.