Meta just announced it will install tracking software on employee laptops to record every keystroke, mouse movement, and screen activity. The reason? To train AI models. The tool is called Model Capability Initiative, or MCI, and it cannot be opted out of.
Let that sink in for a moment.
Your employer isn’t just watching what you produce anymore. They’re watching how you produce it. Every click. Every shortcut. Every weird way you drag files around or tab between apps. And they’re feeding all of it into an AI so it can learn to do your job.
This isn’t a Black Mirror episode. This is a memo from Meta’s CTO Andrew Bosworth, posted this week.
What Exactly Is Meta Doing?
According to TechCrunch and CNBC, Meta is rolling out MCI across U.S. employees’ work computers. The software captures mouse movements, click patterns, keystrokes, and takes periodic screenshots. It tracks activity across hundreds of websites and apps, including Google, LinkedIn, Wikipedia, GitHub, Slack, and Atlassian, along with Meta’s own properties.
Meta says the goal is to train AI agents that can perform workplace tasks autonomously navigating dropdown menus, filling out forms, switching between applications. Their spokesperson said the models need “real examples of how people actually use computers.”
What Meta didn’t say out loud, but what everyone can read between the lines: we’re teaching AI to replace the people doing this work. And the people doing the work are the ones training their replacement. Oh, and Meta is also planning to lay off about 10% of its global workforce starting in May.
The timing isn’t awkward at all.
The Cake Question
Here’s where it gets interesting and where existing employment law starts to feel like it was written for a different planet.
Let’s say I hire you to bake me a chocolate cake. We agree on a price. You make the cake. I get the cake. Simple.
But do I also get the recipe? Do I own the specific way you fold the batter? The trick you use with the oven temperature? The fact that you let it cool for exactly 11 minutes before frosting?
Most people would say no. You paid for the cake, not the baker’s brain.
But that’s exactly what’s happening now. Companies have always owned the output of employee work, the code, the designs, the reports. That’s the cake. Employment law is clear on this: work created within the scope of employment belongs to the employer.
What’s new and what Meta’s MCI program makes painfully obvious is that companies now want the process. Not just what you built, but how you built it. Not just the answer, but how you think. Not just the decision, but the 14 micro-steps your brain went through to get there.
That’s not the cake anymore. That’s the recipe. And nobody agreed to sell that.
This Isn’t Just About Meta
Let’s not pretend Meta is the only company thinking this way. They’re just the first one to get caught saying it out loud.
The entire AI industry is hungry for training data. Companies have already scraped most of the public internet. They’ve licensed books, news articles, and academic papers. Last week, Forbes reported that old startup archives, Slack messages, Jira tickets, internal emails are being scavenged and sold as AI training data.
The next frontier is obvious: the way employees work. Their workflows. Their decision-making patterns. Their problem-solving shortcuts. This is the richest, most valuable training data there is, because it captures not just information, but competence.
And right now, there is no law that clearly protects it.
What the Law Currently Says (And Doesn’t Say)
U.S. employment law was not built for this moment. Here’s where things stand:
Work output belongs to the employer. Under the “work for hire” doctrine, copyrightable work created by employees within the scope of their employment automatically belongs to the employer.
Employee monitoring is broadly legal. In the U.S., federal law places no limit on employer surveillance of work devices. As Yale law professor Ifeoma Ajunwa has noted, Meta’s keystroke logging extends the kind of surveillance previously experienced mainly by delivery drivers and gig workers to white-collar professionals.
But work methods are a gray area. The law is clear that an employer owns the report you wrote. It’s much less clear that they own the way you organize your thoughts before writing it, or the unique sequence of tools you use to research a topic, or the mental model you apply when solving problems. These are closer to personal skills and tacit knowledge, things an employee carries with them from job to job.
Europe sees it differently. EU law, including GDPR, would likely prohibit the kind of monitoring Meta is implementing. Several legal experts have said MCI would be illegal in Europe, which is why the program is currently restricted to the United States.
The Real Problem: You’re Training Your Replacement
Let’s call this what it is.
When Meta captures how an employee navigates a complex Salesforce workflow, or how a developer jumps between GitHub and Stack Overflow to debug a problem, or how a project manager structures a Slack thread to resolve a conflict, they’re not just “collecting data.” They’re extracting the human expertise that makes those employees valuable.
And once the AI learns that expertise, the employee becomes less valuable. Maybe expendable.
This is the uncomfortable math: Your salary pays you to produce work. But the way you produce that work, your efficiency, your instincts, your shortcuts that’s what makes you worth that salary. If a company can capture all of that and embed it in an AI, they’ve essentially cloned your professional value without your consent.
Multiple Meta employees described the MCI program as “dystopian” in internal messages. One wrote that it feels like “everything we do is now going into a black box.” Another simply asked how to opt out. The answer: you can’t.
The Chocolate Cake Framework
So how do we think about this? I keep coming back to the cake analogy because it makes the problem so clear.
There are three layers to any work:
Layer 1: The Output. The cake. The code. The report. The design. This belongs to the employer. Always has. No argument here.
Layer 2: The Process. The recipe. The workflow. The sequence of tools and steps used to create the output. This is where it gets murky. If you developed a unique process on company time using company tools, the company has a reasonable claim.
Layer 3: The Intuition. The baker’s instinct for when the batter is right. The developer’s gut feeling about where the bug is. The manager’s sense for when a team is about to fall apart. This is tacit knowledge, the kind you can’t write down in a manual. It lives in your head. It took years to develop. And no employer should have the right to extract and replicate it without explicit consent and fair compensation.
The problem with Meta’s MCI and with any AI training program that captures employee behavior is that it doesn’t distinguish between these layers. It hoovers up everything. The output, the process, and the intuition all get flattened into “training data.”
What Needs to Change
We don’t have to accept this as inevitable. Here are some principles that could help protect workers while still allowing companies to benefit from AI:
Consent must be real, not buried in onboarding paperwork. Meta says employees agree to monitoring when they’re onboarded. But agreeing to “your work device may be monitored” and agreeing to “your every keystroke will be used to train AI that might replace you” are vastly different things. AI-specific data collection should require separate, explicit, informed consent.
Distinguish between monitoring and mining. Monitoring for security and compliance is one thing. Mining behavioral data to train AI models is something else entirely. Companies need to be transparent about which they’re doing and why.
Employees should have rights over their behavioral data. Just as GDPR gives European citizens rights over their personal data, workers should have the right to know what behavioral data is being collected, how it’s being used, and the right to say no, without fear of losing their job.
Compensation for AI training contributions. If an employee’s work patterns are valuable enough to train an AI model, they’re valuable enough to compensate. Companies could offer AI training bonuses, royalty arrangements, or equity participation when employee behavioral data is used in model development.
Portable skills must stay portable. The skills, techniques, and problem-solving approaches an employee brings to a job and takes to the next one should not be captured and locked into a proprietary AI model. Employment agreements should explicitly carve out tacit knowledge and transferable skills from any IP assignment.
Sunset clauses on behavioral data. Behavioral data collected from employees should have expiration dates. If an employee leaves, their behavioral data should be removed from training datasets within a reasonable period, similar to how GDPR handles data deletion requests.
Where This Is Headed
Meta is spending up to $135 billion on AI in 2026. They acquired a 49% stake in Scale AI for over $14 billion. They’re building what they call “Meta Superintelligence Labs.” And now they’re turning their own employees into training data.
This isn’t going to stay a Meta story. Every large company with AI ambitions is going to face the same temptation: why pay humans to do work when you can also capture how they do it and teach a machine?
The companies that handle this well with transparency, consent, and fair compensation will attract and retain the best talent. The ones that don’t will find themselves in a race to the bottom, where the smartest employees either leave, game the system, or simply stop being creative on company time.
Because here’s the thing Meta and others might not have considered: if employees know their every move is being captured to train an AI, they’ll stop showing you their best tricks. They’ll follow the manual. They’ll click the obvious path. They’ll be perfectly average.
And the AI trained on that data? It’ll be perfectly average too.
You can monitor the recipe. But you can’t force someone to share their secret ingredient.
Sources: TechCrunch (April 21, 2026); CNBC (April 22, 2026); Fortune (April 21, 2026); Reuters (April 21, 2026); Biometric Update (April 22, 2026).
Originally published on https://aibigenie.com/.