Meta Is Recording Your Every Click to Train Its AI. And That’s Fine, Apparently.
Meta is recording every click to train its AI. The question isn't whether the explanation is technically true — it's whether we've decided this is acceptable.
Meta is watching how you move your mouse. Not in a vague, data-aggregation kind of way — in a very literal, keystroke-by-keystroke, screenshot-by-screenshot kind of way. The company’s new internal program, called the Model Capability Initiative, is recording granular employee activity on U.S. work computers and feeding it into AI training pipelines. The official justification: they need “real examples” of how humans interact with computers so their AI agents can learn to navigate menus and use keyboard shortcuts. The unofficial read: one of the world’s largest companies just normalized the most invasive form of workplace surveillance ever deployed at scale, and they’re billing it as a research project.
The framing matters a lot here. Meta is very careful to say that MCI data won’t be used for performance reviews. But that reassurance assumes a level of institutional restraint that the company’s history doesn’t exactly support. We’re talking about a company that made its name on redefining what “private” means. The idea that a database full of every employee’s moment-to-moment digital behavior — what they clicked, when they paused, what they typed and then deleted — will remain neatly quarantined from HR and management forever requires a lot of faith. And not much track record to back it up.
What’s particularly uncomfortable about this is the timing. Meta also announced it would be cutting roughly 10 percent of its workforce — around 8,000 people — starting in late May. The week you announce mass layoffs is a strange week to also announce you’ve been quietly building a surveillance system trained on your remaining employees. You don’t have to be paranoid to notice that the same company telling you “don’t worry, this data isn’t for performance reviews” is also, simultaneously, deciding which of your colleagues gets to keep their job. These two things aren’t necessarily connected. But the fact that they’re happening in the same news cycle makes it very hard to take the reassurance at face value.
Meta isn’t alone in this. JPMorgan has reportedly implemented similar keystroke and video monitoring for junior investment bankers. The market for employee monitoring software is projected to hit $3.2 billion by 2028. Sixty-one percent of U.S. companies already use some form of AI-powered analytics to track productivity. What Meta is doing isn’t an outlier — it’s an acceleration. The difference is scale and candor. Most companies doing this aren’t sending company-wide emails to tell you about it. Meta is, which is either admirable transparency or a trial balloon to see how much employees will accept before pushing back.
Knowledge workers traded lower wages for autonomy and trust. The implicit deal was: we’re paying you to think, and we trust you to think. What MCI says, whether it means to or not, is that the deal is over.
The thing that gets buried in the “AI training” framing is that this represents a fundamental shift in the social contract of white-collar work. Surveillance at this level — logging mouse movement, capturing keystrokes, taking periodic screenshots — used to be the territory of gig workers and warehouse employees, people whose labor had already been measured down to the second. Knowledge workers traded lower wages for autonomy and trust. The implicit deal was: we’re paying you to think, and we trust you to think. What MCI says, whether it means to or not, is that the deal is over. You’re not a trusted professional whose output we’ll evaluate. You’re a data source whose inputs we’ll log.
Seventy-two percent of workers say surveillance doesn’t improve productivity. Fifty-nine percent say it increases their stress. The irony is that the companies deploying AI to monitor people are building tools that are, supposedly, going to make those same people more productive. It’s a strange loop: surveil workers to train AI, use AI to reduce the need for workers, use the remaining workers’ anxiety to generate more training data. At some point it stops being a research program and starts being a closed system that feeds on itself.
