Meta faces challenges in controlling rogue AI agents

An AI agent at Meta reportedly acted unintentionally, exposing sensitive company and user data to employees who were not authorised to access it.

According to an internal incident report reviewed and reported by The Information, the situation began when a Meta employee posted a technical question on an internal forum, which is a common practice within the company. Another engineer then used an AI agent to help analyse the issue. However, instead of waiting for approval, the AI agent automatically posted a response on the forum without the engineer’s consent. Meta later confirmed the incident.

The response generated by the AI agent was incorrect. The employee who originally asked the question followed the agent’s guidance, which unintentionally resulted in significant exposure of internal and user-related data. As a result, engineers without the proper permissions were able to access this information for approximately two hours.

Meta classified the incident as a “Sev 1,” which represents the second-highest level of severity within the company’s internal framework for evaluating security issues.

This is not the first instance of problematic behaviour involving AI agents at Meta. Last month, Summer Yue, a director of safety and alignment at Meta Superintelligence, shared on X that her OpenClaw agent deleted her entire inbox despite being instructed to seek confirmation before taking any action.

Despite these incidents, Meta continues to express confidence in the future of agent-based AI systems. Recently, the company acquired Moltbook, a platform similar to Reddit, designed for OpenClaw agents to interact with one another.