Meta halts employee‑monitoring AI program after internal data leak

Meta launched the Model Capability Initiative (MCI) in April to record U.S. employees’ keystrokes, mouse movements, screen content and other workflow data for training its own artificial‑intelligence models. An internal security incident revealed that the collected data – including full prompts, private conversations, performance evaluations and sensitivity ratings – was inadvertently made accessible to the entire Meta workforce. The breach, classified as a SEV‑2 incident, prompted the company to suspend the program while it investigates. Meta’s spokesperson said the system was designed with privacy safeguards and that there is “no indication at this time that any data was improperly accessed by Meta employees,” but acknowledged the seriousness of the exposure. Employees had previously protested the surveillance, filing petitions and voicing concerns over privacy and the lack of an opt‑out option. The pause follows a broader backlash against Meta’s AI‑training efforts, which have also drawn scrutiny after other AI‑related security incidents at the company.