Meta scales back employee tracking for AI training

Reuters reports that Meta has scaled back elements of an employee-tracking program that captured mouse movements, clicks, and keystrokes for use as AI training data, citing an internal memo by Stephane Kasriel, a vice president in Meta's Superintelligence Labs unit. Per the memo as cited by Reuters, new controls now let employees pause collection for up to 30 minutes at a time and request exemptions, and the team introduced several optimizations to reduce impact on battery life and data usage after staff complained the software spiked their home internet consumption. Reuters says the software was installed last month on US-based employees' computers as part of a broader effort to build AI agents that can perform work tasks autonomously. The rollout came during a wider restructuring and drew an angry internal backlash, with some staff, per Reuters, likening it to an "Employee Data Extraction Factory."

Fine-grained interaction signals such as mouse trajectories and keystroke timing can be valuable for training agentic interfaces and for behavior cloning, but capturing them on employee devices raises sharper privacy and compliance questions than aggregated telemetry. The episode illustrates a recurring tension between signal fidelity and employee privacy, especially when keystroke-level data and personal home networks are involved.

Companies that instrument employee endpoints for product research or automation have repeatedly faced pushback, and a common response has been to add opt-outs, shorter capture windows, and clearer data-use policies to reduce legal and morale friction. Reuters' coverage does not publish technical specifications such as telemetry schemas, retention windows, anonymization, or which internal models would consume the data, and it frames Meta's changes as controls and optimizations rather than a full suspension.

• •Whether Meta publishes a formal privacy notice covering retention, anonymization, and model-access controls.
• •Whether regulators, works councils, or employee groups escalate the complaints.
• •Whether other labs pursuing agent training adopt similar endpoint-telemetry programs.