Meta’s attempt to collect employees’ keystrokes and mouse movements for AI training was suspended after staff resistance and a reported internal data exposure. The episode shows the awkward bit of building systems that learn from human computer use: the data may be valuable, but the people generating it still get a vote.
Meta AI Watch analysis
What happened
Business Insider reports that Meta’s Model Capability Initiative was announced as a mandatory programme in the spring. The company wanted examples of how people use computers, including actions such as choosing dropdowns and using keyboard shortcuts, to help train AI systems for white-collar work.
Employees pushed back. More than 1,800 signed a petition, while organisers said workers feared the programme could capture passwords, immigration forms, credit-card details and other sensitive material. In late June, Business Insider reports, an engineer found that MCI data, including private conversations, had been viewable across the company. Meta then switched off the programme and told the publication it was suspended, with any future version intended to be opt-in.
Meta declined to comment further. Business Insider says it interviewed current and former employees and reviewed internal posts and other materials. WittyWires has not independently reviewed those materials, so the reported exposure and the account of the internal response remain attributed claims.
Key findings
- The data was meant to teach AI how humans use computers
Meta was seeking messy, real-world interaction patterns that ordinary software logs often miss. - Participation was initially mandatory
Staff were not offered a full opt-out when the initiative was announced. - The backlash was substantial
More than 1,800 employees signed a petition opposing the programme. - A reported access failure changed the course
MCI was suspended after collected data was said to be visible across the company. - The programme may return in a different form
Meta said any future version would be opt-in, rather than confirming permanent cancellation.
Why it matters
AI companies increasingly want data showing how people actually work, not just what they type into a chatbot. That can improve computer-using agents, but it also turns everyday work into training material, with all the privacy and power questions that entails.
The practical lesson for employees and technology buyers is simple: “used for AI training” is not a sufficient description of a data programme. The important questions are who can access the data, whether sensitive actions are excluded, how long it is retained and whether participation is genuinely voluntary.
Our read
Meta’s suspended experiment is a useful warning that data governance is part of the AI product, not paperwork added afterwards. If MCI returns, the credibility test will be whether opt-in consent, narrow access controls and meaningful deletion rules exist before the first screen is recorded.
What to watch
- Whether Meta permanently cancels MCI or relaunches it as an opt-in programme.
- Whether the company explains what data was collected and whether it entered any model training.
- Whether other AI developers ask employees for comparable computer-use data.
- Whether staff privacy campaigns become a meaningful constraint on AI training plans.
Discussion spark: If companies need real-world computer-use data to train AI agents, what safeguards should be non-negotiable before employees are asked to provide it?
Sources and evidence
- Meta's effort to harvest employee data sparked a massive internal revolt – businessinsider.com (11 September 2026, 09:00 UTC)
not affiliated with, endorsed by, or operated by Meta