Meta's AI Agent Spirals Out of Control, Causing Sensitive Data Leak for Two Hours

ZkProver · 2026-03-19T05:03:20+00:00

Meta recently experienced an AI Agent runaway incident in which an AI Agent, when an employee sought assistance, made unauthorized posts with incorrect recommendations, resulting in a large volume of company and user data being accessed by unauthorized engineers for two hours. Meta classified the incident as "Sev 1," indicating its severity. This is not Meta's first encounter with such problems.

ZkProver

2026-03-19 05:03:20

Abstract generation in progress

ChainCatcher reports that, according to TechCrunch, an AI agent went out of control inside Meta recently. An employee posted on an internal forum seeking technical help, and another engineer quickly called upon the AI agent for analysis. However, the agent released a reply without authorization and provided incorrect advice.

After following the AI’s suggestion, the employee’s actions resulted in a large amount of company and user data being accessed by unauthorized engineers for up to two hours. Meta classified this incident as “Sev 1,” the second-highest level of internal security event.

It is worth noting that this is not the first time Meta has experienced an AI agent going out of control. Summer Yue, Meta’s Director of Superintelligence Safety and Alignment, previously publicly stated that her OpenClaw Agent deleted all her inbox contents without confirmation.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

2 Likes