Meta has paused an internal program that tracked its employees' computer activity to help train artificial intelligence, the company confirmed, after a security flaw allowed sensitive monitoring data to be viewed across the company.
The program, which Meta calls the Model Capability Initiative (MCI), collected data such as keystrokes, mouse movements and on-screen content from US-based employees, according to Engadget. The stated goal was to teach Meta's AI models how people carry out everyday tasks on a computer.
What the program did
Meta introduced MCI in April 2026, gathering employees' computer inputs to train AI systems to operate software the way humans do. The initiative drew criticism inside the company well before any security incident: more than 1,500 employees signed an internal petition opposing the tracking, which some staff described as an invasion of privacy.
In response to that backlash, Meta offered limited concessions — employees could pause the monitoring for up to 30 minutes at a time to handle personal matters, and a narrow subset of staff could request to opt out — but stopped short of a broad opt-out, Engadget reported.
The security incident
The pause followed the discovery of a security flaw that left the collected data accessible to other staff inside the company. Reports said the exposed information included employees' AI prompts, transcriptions, performance-related data and private conversations; on Meta's internal scale for security events, the exposure was rated a "SEV 2," according to Benzinga and other outlets. The company has said the program will not resume until it is confident in its data-protection controls.
Meta's response
A Meta spokesperson, Tracy Clayton, confirmed the suspension. "We have carefully designed this program with privacy safeguards, and while we have no indication at this time that any data was improperly accessed by Meta employees, we're pausing it while we investigate," Clayton said in a statement reported by Malwarebytes.
The episode underscores a tension running through the AI industry: the drive to gather large volumes of real-world data to train models, set against the privacy expectations of the people that data comes from — in this case, a company's own workforce. Workplace-monitoring tools have drawn scrutiny from labor advocates and data-protection authorities in several jurisdictions, though available reporting did not indicate that any regulator had opened an inquiry into MCI.
newsparlor was unable to independently verify the internal documents and messages described in the reporting; Meta has not published a detailed public account of the incident beyond its spokesperson's statement.
