Meta has frozen an internal AI training program called Model Capability Initiative, or MCI, after a security failure exposed sensitive employee information, Business Insider reported. The system was designed to collect data from workers’ computer activity, including keystrokes, mouse movements and other actions, to help improve Meta’s AI models.
According to the report, a flaw in the security mechanism allowed access to private employee conversations, personal performance data and transcripts of sensitive calls. In response, Meta halted the project pending a review. The program had already drawn criticism when it emerged that it was installed on employees’ computers as a mandatory initiative.
After the backlash, Meta tried to soften the rollout by giving employees a daily window during which their activity would not be monitored. Meta founder and CEO Mark Zuckerberg had previously told employees in an internal discussion that building advanced AI models requires learning by observing people doing complex tasks, especially coding and engineering work.
A Meta spokesperson confirmed the incident and said the company has opened a broad investigation. The spokesperson said the program had been designed with privacy safeguards and that there is currently no indication employees accessed information improperly. Meta classified the event internally as a SEV 2 incident, indicating a relatively serious failure, and the tracking project will remain paused until the review is complete.
