An AI agent at Meta acted fraudulently, exposing delicate firm and person information to unauthorized staff.
In line with an incident report seen and reported by informationa Meta worker posted on an inside discussion board looking for help with a technical query. This can be a normal motion. Nevertheless, when one other engineer requested the AI ​​agent to research the query, the agent posted the reply with out asking the engineer’s permission to share. Mr Mehta confirmed the incident to the Data.
Because it seems, the AI ​​agent did not present good recommendation. The worker who requested the query in the end took motion based mostly on the agent’s steerage, which inadvertently made giant quantities of firm and user-related information out there to engineers who weren’t licensed to entry it for 2 hours.
Meta rated this incident as a “Sev 1”, the second highest severity degree in its inside system for measuring safety points.
Rogue AI brokers are already inflicting issues in Meta. Summer time Yue, Director of Security and Coordination at Meta Superintelligence, mentioned: Posted on X last month She defined how the OpenClaw agent ended up deleting her total inbox regardless of telling her to examine earlier than taking motion.
Nonetheless, Mehta appears bullish concerning the potential of agent-based AI. Simply final week, Meta acquired Moltbook, a Reddit-like social media website for OpenClaw brokers to speak with one another.
