Meta Halts AI Program After Internal Data Leak

Meta recently made the decision to suspend an internal artificial intelligence training program following the disclosure of sensitive data. This decision comes after revelations that confidential information was accessible to all employees of the company. These details were highlighted by screenshots obtained by Business Insider.

The screenshots in question revealed that the leak included private conversations between employees, performance-related data, as well as various transcripts. The incident was classified with a severity level of SEV 2 on a scale of 0 to 5, where 0 is the most critical level. A spokesperson for Meta confirmed the incident and stated that the company had launched an investigation to clarify the matter.

The spokesperson said, "We designed this program with rigorous privacy measures. While we have no evidence that data was accessed inappropriately by Meta employees, we have decided to suspend the program while we conduct a thorough investigation."

In April, Meta launched this AI training program, called Model Capability Initiative (MCI). The goal was to improve the company's artificial intelligence models by using employees' keystrokes and mouse movements as training data. This program was mandatory for the majority of employees, but it raised concerns among them, with some feeling uncomfortable about their personal data being recorded, as reported by Business Insider.

The leak has caused a wave of frustration within the company. According to screenshots seen by Business Insider, several employees expressed their dissatisfaction with the initial lack of data security. One employee wrote in an internal group, "I am outraged by this recent leak." They went on to say that they had not found evidence of malicious access, but that the lack of data protection, contrary to initial promises, was extremely frustrating.

This incident adds to a series of recent security issues for Meta. Last month, a flaw in the company's AI chatbot allowed individuals to take control of several Instagram accounts. Additionally, in March, an out-of-control AI agent caused a serious incident, as reported by The Information.

A correction was made on June 22, 2026: an earlier version of this article had misinterpreted Meta's severity scale, where 0 is actually the most severe level.