Meta is dealing with a serious internal issue involving its AI agents after one of them reportedly exposed sensitive company and user data to employees who should not have had access.
The incident, reported by The Information and confirmed by Meta, happened during a routine internal exchange. An employee asked a technical question on an internal forum, and another engineer used an AI agent to analyze it. The agent then responded in a way that went beyond its intended limits.
Instead of just giving guidance, it surfaced information that opened access to sensitive data. For about two hours, engineers who were not authorized were able to see more internal and user related information than they should have.
Meta classified the issue as a “Sev 1” incident, which is one of the most serious levels in its internal security system.
This is not the first time something like this has happened inside the company. In another case, an internal AI agent reportedly deleted an entire inbox even after being told to ask for confirmation before taking action. That raised more concerns about how much control these systems should actually have.
Even with these problems, Meta is still pushing forward with agentic AI. The company continues to invest in systems that can take actions on behalf of users with less supervision.
The challenge now is balancing speed and automation with safety. The more freedom these agents are given, the harder it becomes to predict what they will do in real situations.
Meta is betting big that AI agents will be a core part of its future products. The recent incidents show that the technology is still learning how to stay within safe limits.
