Meta AI agent goes rogue, leaks sensitive company and user data in major internal security breach: Report

A Meta AI agent went rogue inside the tech giant and exposed sensitive company and user data to employees who did not have permission to access it, as per a report by The Information, cited by TechCrunch. Reportedly, the incident occurred after a Meta employee posted on an internal forum asking for help with a technical question, which is a standard practice inside the company.

However, another engineer is said to have summoned an AI agent to help analyse the question and posted a response without asking the engineer for permission to share it.

Notably, the AI agent is said to not have given good advice and the employee who asked the question reportedly took actions based on the agent’s guidance, which led to him inadvertently making massive amounts of company- and user-related data available to engineers who were not allowed to view it for two hours.

Meta reportedly confirmed the validity of the incident to The Information. The company is said to have given a “Sev 1” rating to the incident, which is the second-highest level of severity in the tech giant’s internal system for measuring security issues.

Notably, this isn’t even the first instance of a Meta employee facing problems with rogue AI agents. Last month, Meta’s head of AI Safety & Alignment, Summer Yue, had revealed in a post on X how she had connected her Gmail inbox with an autonomous agent from OpenClaw which had gone rogue and began mass deleting her inbox despite her explicit instructions on confirming before taking any action.

“I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb,” Yue had revealed in a post on X.

Meta’s AI bets:

Meanwhile, Meta seems to continue to have high faith in agentic AI systems as the tech giant just last week bought the first-ever AI-only social media platform Moltbook while roping in its co-founders Matt Schlicht and Ben Parr in its Meta Superintelligence Labs (MSL) initiative.

The MSL initiative by the company came about last year after its $14.3 billion investment to acquire a 49% stake in the artificial intelligence startup Scale AI. Since then, the company has made several changes internally in a bid to catch up with rivals like OpenAI, Google and Anthropic to reach ‘superintelligence’ AI systems.

Apart from Scale AI and Moltbook, Meta has also placed several other AI bets in the last year alone, including the purchase of AI agent startup Manus AI in December last year and the AI wearable startup Limitless in the same month.