Meta has paused its Model Capability Initiative, an AI training program that tracked employee keystrokes and mouse movements, after a data leak exposed sensitive information. The pause follows an internal error where a researcher inadvertently moved transformed employee data to an unauthorized location, according to Meta CTO Andrew Bosworth.
Core Facts
Meta’s AI training program, introduced in April 2024, was designed to improve AI models by monitoring employee activity. However, the program faced widespread backlash due to its mandatory participation policy and privacy concerns. In June 2024, the program was paused after a leak made employee data accessible to the entire company, according to Business Insider. Bosworth confirmed the data was moved internally by a researcher but denied any breach or foul play.
Deeper Context
Program Details and Backlash
The Model Capability Initiative involved installing software on most U.S. employees' devices to collect keystrokes and mouse movements. Meta maintained the program had privacy safeguards, but employees could not opt out, leading to low morale, as Bosworth acknowledged in an internal meeting. The program was paused indefinitely while Meta investigates the incident.
Official Statements
Bosworth stated the data was secure and only accessible to a small group before the error. He emphasized the company was locking down the program to prevent further issues. A Meta spokesperson confirmed the pause but noted no evidence of improper data access by employees.
Employee Reactions
The program’s mandatory nature and lack of transparency contributed to employee dissatisfaction. Bosworth described morale as among the worst in Meta’s history, though no formal employee statements were provided in the sources.
Investigation and Next Steps
Meta has not specified a timeline for resuming the program. The investigation focuses on preventing future data misplacement and ensuring compliance with internal protocols.