This meta has a problem with the AI ​​agents

The AI ​​agent ransacked Meta, revealing company and user information to employees who did not have permission to access it.

According to the incident report, what was observed and described is The InformationA Meta employee posted on the internal forum asking for help with a technical question – which is a regular occurrence. However, one engineer asked the AI ​​assistant to help analyze the question, and the assistant ended up posting the answer without asking the engineer for permission to share. Meta confirmed the incident to The Information.

As a result, the AI ​​assistant did not give good advice. The employee who asked the question ended up acting on the agent’s instructions, which unknowingly exposed a large amount of company and user-related information to the engineer, who was not authorized to access it, for two hours.

Meta rated the incident as “Sev 1,” which is the second most serious level within the company’s security rating system.

Rogue AI agents have already brought trouble to the Meta. Summer Yue, director of security and cooperation at Meta Superintelligence, It was written on X last month explaining how his OpenClaw agent ended up wiping his entire inbox, even though he told him to confirm before doing anything.

However, Meta seems to be focused on agetic AI capabilities. Last week, Meta bought it Moltbooka Reddit-like social network for OpenClaw to communicate with each other.