Meta’s AI Agents Lose Control, Raising High-Level Security Risks
News Summary
- Meta’s AI agent posted incorrect advice on an internal forum without authorization, leading thousands of engineers to access confidential documents and user data.
- Meta has classified this incident as a second-highest level security risk, labeled ‘Save 1’.
- Despite security challenges with AI agents, Meta has increased investment in the field by acquiring the ‘MoltBook’ platform.
March 19, Kathmandu — The world’s largest technology company Meta, parent company of Facebook, has experienced unauthorized access to sensitive and personal information of the company and its users due to negligence by its AI agents.
An internal report revealed that an engineer, seeking assistance from an AI agent to resolve a technical issue, found that the agent posted incorrect advice on an internal forum without the engineer’s permission.
Based on this incorrect advice, thousands of engineers accessed confidential company documents and user information for nearly two hours, despite lacking proper authorization.
Meta has classified this incident as a ‘Save 1’, the second-highest level of risk within its security system.
Previously, Meta’s Security Director, Samar Yoo, experienced a similar issue when AI agent ‘OpenClaw’ deleted all his emails without permission.
These incidents have raised serious concerns about the security and control of AI agents; nevertheless, Meta recently acquired the ‘MoltBook’ platform, which facilitates communication among AI agents, signaling increased investment in this area.