Meta Is Training AI by Tracking Worker Movements — Keystrokes, Mouse Clicks, and Screenshots From Employees' Own Computers
AI / AI Tech Trends | 4 min read
Meta is installing new tracking software on US-based employees' computers to capture mouse movements, clicks, keystrokes, and occasional screenshots — and feeding that data directly into its AI training pipeline. The tool, internally named Model Capability Initiative (MCI), was disclosed in a memo posted by a staff AI research scientist in an internal channel for the company's model-building Meta SuperIntelligence Labs team. The purpose, according to the memo, is to improve Meta's AI models in areas where they still struggle — such as choosing from dropdown menus, navigating basic computer interfaces, and using keyboard shortcuts. MCI will run on a designated list of work-related apps and websites and take occasional snapshots of screen content for context. Employees were told they can contribute simply by doing their daily work.
"If we're building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them — things like mouse movements, clicking buttons, and navigating dropdown menus."
— Andy Stone, Spokesperson, Meta
The Closed Loop Vision — Agents That Do the Work, Humans That Direct and Improve
The MCI programme reflects a broader strategic objective at Meta: building AI agents capable of performing white-collar work tasks autonomously. CTO Andrew Bosworth described the goal in a memo as "a closed loop" in which agents could "automatically see where we felt the need to intervene so they can be better next time." His framing: "The vision we are building towards is one where our agents primarily do the work and our role is to direct, review and help them improve." Meta has stated the data collected via MCI will not be used for performance assessments — only for model training — and that safeguards are in place to protect "sensitive content," without specifying which types of data would be excluded. The company has been internally exhorting staff to use AI agents for coding and other tasks — even if it slows them down in the short term — and recently created a new Applied AI (AAI) engineering team aimed at improving the coding capabilities of Meta's AI models and building AI agents that can perform the bulk of the work to build, test, and ship future products and infrastructure. Meta has also moved to replace certain job functions with a new general-purpose title: "AI builder."
Layoffs, $140 Billion in AI Spend, and a Workforce in Transition
The MCI initiative arrives as Meta is navigating one of the most significant workforce restructurings in its history. The company is planning to lay off 10% of its global workforce starting 20 May 2026, having already cut approximately 2,000 employees in smaller rounds — with job listings on its careers website falling sharply from around 800 in March to just seven. Internally, Mark Zuckerberg previously stated that 2026 would be "the year that AI dramatically changes the way we work." Meta is projected to spend approximately $140 billion on AI in 2026 — nearly double the previous year. The company acquired a 49% stake in data-labelling firm Scale AI for more than $14 billion, with Scale's former CEO Alexandr Wang now leading Meta SuperIntelligence Labs. Employee reactions have been mixed: one worker, speaking anonymously, described the MCI tracking as "very dystopian", particularly amid expectations of further job cuts. Another former staff member described it as "just the latest way they're shoving AI down everyone's throat."
Legal and Regulatory Questions — US, Europe, and GDPR
The legality of MCI varies significantly by jurisdiction. In the US, there is currently no federal limit on worker surveillance — with state-level laws requiring at most that workers be broadly informed when monitoring is taking place. However, the picture is markedly different across Europe. Valerio De Stefano, a law professor at York University in Toronto who studies technology and comparative labour law, said the practice would likely be considered a violation of Europe's General Data Protection Regulation (GDPR). In Italy, using electronic monitoring to track employee productivity is explicitly illegal; in Germany, courts have held that employers can deploy keystroke logging only in exceptional circumstances, such as suspicion of a serious criminal offence. Ifeoma Ajunwa, a scholar on worker surveillance, noted the sharp contrast between US permissiveness and the protections that exist in other jurisdictions — a tension that is likely to intensify as AI-driven workplace monitoring becomes more widespread.
Key Takeaways
- • Meta is installing the Model Capability Initiative (MCI) — tracking software on US-based employees' computers capturing mouse movements, clicks, keystrokes, and occasional screenshots — to train AI models. Disclosed via memo by a staff AI research scientist in the internal Meta SuperIntelligence Labs team channel. Runs on a designated list of work-related apps and websites. Employees contribute by doing their normal daily work. Data will not be used for performance assessment — model training only. Safeguards in place for "sensitive content."
- • Strategic objective: build AI agents capable of performing white-collar work tasks autonomously. MCI addresses a gap in AI model capabilities — replicating basic computer-use behaviours like dropdown navigation and keyboard shortcuts. CTO Andrew Bosworth's vision: a "closed loop" where agents automatically identify where human intervention occurred and improve accordingly. The broader aim: agents primarily do the work, while humans direct, review, and improve.
- • Workforce context: Meta is planning a 10% global workforce reduction starting 20 May 2026, having already cut ~2,000 in smaller rounds. Job listings dropped from ~800 in March to 7. Meta is spending ~$140 billion on AI in 2026 (nearly 2× 2025). 49% stake in Scale AI acquired for $14B+; Alexandr Wang leads Meta SuperIntelligence Labs. New Applied AI (AAI) engineering team created to improve AI coding capabilities. New general-purpose "AI builder" job title replacing certain function-specific roles. Zuckerberg: 2026 is "the year that AI dramatically changes the way we work."
- • Employee reaction: one anonymous worker described MCI as "very dystopian," particularly in the context of expected further layoffs. Another former employee called it "just the latest way they're shoving AI down everyone's throat." The combination of AI-driven workforce surveillance for model training alongside active layoffs — and the replacement of job titles with "AI builder" — has created significant internal tension at the company.
- • Legal and regulatory landscape: US federal law places no limit on worker surveillance; state laws require only broad disclosure. In Europe: GDPR would likely prohibit such monitoring (per Valerio De Stefano, York University Toronto); Italy explicitly prohibits electronic monitoring for productivity tracking; Germany permits keystroke logging only in exceptional circumstances (e.g. suspicion of serious crime). MCI is US-only in its current deployment — but as Meta operates globally, the broader question of whether similar data collection could ever extend to European operations represents a significant legal constraint. The broader industry pattern: OpenAI was reported requesting contractors to upload real work products for training; Meta acquired Scale AI for labelling infrastructure. AI training data hunger is reaching directly into the workplace itself.