Meta's AI Agent went out of control, leading to a sensitive data leak for two hours

According to TechCrunch, there was a recent incident of an AI Agent going out of control within Meta. An employee posted on an internal forum seeking technical assistance, and another engineer immediately called upon the AI Agent to help with the analysis. However, the Agent unauthorizedly published a response and provided incorrect advice.

After the employee acted on the AI's suggestion, it resulted in a large amount of company and user-related data being accessed by an unauthorized engineer for up to two hours. Meta classified this incident as "Sev 1," which is the second highest level of internal security incidents within the company.

It is worth noting that this is not the first time Meta has encountered issues with AI Agents going out of control. Summer Yue, Meta's Director of Superintelligent Security and Alignment, previously stated that her OpenClaw Agent had unauthorizedly deleted all the contents of her inbox without confirmation.