Bad idea, bad timing: Meta wanted to monitor every click of its employees for its AI and they said no

The employees of Meta succeeded in forcing their management in the United States to bend. Since April, software installed on their work computers recorded mouse movements, clicks, keystrokes and screenshots. All this in order to train the company's AI agents.

The decision really did not please the employees concerned. And on June 2, after several weeks of rebellion, Meta backed down.

The program is called the Model Capability Initiative, or MCI. According to Reuters, Meta announced, in an internal memo signed Stephane Kasriel (vice-president of the Superintelligence Labs division), that it had scaled back (without canceling the project). Concretely, employees will now be able to pause data collection in 30-minute increments, and only certain profiles will obtain a complete exemption.

Teaching AI to click like a human

The initial idea is not absurd on paper. Meta's AI agents, these software programs supposed to carry out tasks on their own, still struggle with basic gestures (dropping down a menu, using a keyboard shortcut, navigating in an interface). To progress, they need real examples of how humans use computers. Hence the idea of ​​watching (spying?) on employees working.

Andrew Bosworth, CTO of Meta, sums up the vision bluntly: ultimately, agents will do most of the work, with humans simply supervising them.

Where it gets stuck is the timing and the method. The software was deployed in full social plan: Meta confirmed the elimination of around 10% of its workforce, or 8,000 positions. Asking employees to train AI that could one day replace them, while eliminating thousands of jobs... The message doesn't come across well.

In Europe, the project would undoubtedly be illegal

Note that this device would probably not work in our home. The systematic capture of keystrokes, without any real possibility of objecting, collides head-on with the GDPR and the CNIL doctrine concerning surveillance at work. In France, keyloggers are considered a violation, except in exceptional cases with prior information of employees and consultation of the CSE.

Amazon France Logistique has also paid the price: the company was fined 32 million euros in 2024, reduced to 15 million by the Council of State at the end of 2025 for surveillance practices deemed excessive.

In the United States, the framework is radically different. On an employer-provided device, there are almost no legal limits to this type of surveillance, and the program initially provided no option to opt out. It is precisely this point that crystallized the anger.

Mark Zuckerberg may swear that data is not used to monitor or evaluate performance, but the uneasiness persists (and this was predictable).

Finally, remember that this logic of AI agents working in your place is the direction other tech giants are taking. Gemini Spark from Google and Microsoft Scout are good, very recent examples.