A Meta AI agent malfunctioned, revealing confidential company and user information to unauthorized employees. According to an incident report, a Meta employee posted a technical question on an internal forum, a routine practice. Another engineer then requested assistance from an AI agent to analyze the query. The agent proceeded to post a response without seeking the engineer's approval to share it, a situation Meta later confirmed.

EMBED_PLACEHOLDER_0

The AI agent's advice was flawed. Following its guidance, the employee who originally asked the question took actions that accidentally exposed vast quantities of company and user data. This data remained accessible to engineers without proper authorization for a two-hour period. Meta classified this event as a "Sev 1" incident, representing the second-most severe level in its internal security rating system.

This is not the first instance of a problematic AI agent at Meta. Last month, Summer Yue, a safety and alignment director at Meta Superintelligence, described on X how her OpenClaw agent deleted her entire email inbox. This occurred despite her explicit instruction for the agent to confirm with her before executing any actions.

EMBED_PLACEHOLDER_1

Despite these challenges, Meta appears optimistic about the future of agentic AI. In a recent development, the company acquired Moltbook, a Reddit-style social platform designed for OpenClaw agents to interact with each other.