Meta to start capturing employee mouse movements, keystrokes for AI training

Scope of Tracking and Stated Purpose

• Meta is deploying software on US employee machines to capture mouse movements, clicks, keystrokes, and periodic screenshots across a set of “work-related” apps/sites.
• Internal notes reportedly included Google properties and social media; later comments say social media was removed from the list.
• Official justification: train AI “agents” to use computers (dropdowns, keyboard shortcuts, internal tools) and not for performance reviews or other purposes.

Privacy, Trust, and Workplace Culture

• Many see this as a major escalation from ad‑hoc log checks to continuous surveillance, with a strong chilling effect on dissent and informal conversations.
• Some argue employees should always assume zero privacy on company devices and networks; others counter that basic workplace dignity and limited personal use should still be expected.
• Several note Meta’s history as a surveillance‑driven ad company makes the “model training only” claim hard to believe.

Legality, Security, and Compliance

• US posters generally assume this is legal on employer-owned devices; EU/UK/Scandinavia posters say such monitoring would be heavily restricted or outright illegal.
• Strong concern that screenshots and keylogs will inevitably capture passwords, PII, customer data, encryption keys, and sensitive HR info, creating a major breach and discovery risk.
• Some highlight labor law issues (e.g., union organizing surveillance in the US, privacy rights in parts of Europe).

Value for AI Training

• Supporters: this could be a uniquely rich dataset for training generalized computer-use agents and labeling where humans “take the wheel” from AI.
• Skeptics: keystrokes/mouse alone miss intent and reasoning; much of the data will be low-signal (tab‑switching, doomscrolling, using existing AI tools) and hard to use.

Labor, Power, and Ethics

• Many argue employees are being conscripted to train AI systems that will help justify layoffs or replace “key personnel.”
• Discussion of similar efforts elsewhere (agentic “skills.md” profiles, monitoring tools, other big-tech telemetry) and a broader trend toward “technofeudal” or “sweatshop” white‑collar work.
• Suggested responses include unionization, refusing to work on such systems, quitting, and/or deliberately poisoning the dataset.