Meta halts employee tracking for AI training after 1,600 workers sign petition

Meta has paused a program that ran in the background on employees' work computers, tracking keystrokes, mouse movements and screen contents. The data was collected to train the company's own AI models. The pause came after around 1,600 workers signed an internal petition demanding that the program be stopped.

What exactly the tracker recorded

The tool, deployed by Meta on corporate machines, continuously recorded everything an employee did on their computer: every keystroke, every mouse click and everything shown on screen at any moment of the working day. In effect, the program created a detailed, real-time record of each worker's professional activity.

Meta planned to feed the collected data into the training of its language and multimodal AI models. The logic is obvious: live work sessions of experienced specialists contain behavioral patterns and contextual signals that cannot be obtained from public sources.

The startup Adept ran similar experiments, trying to train agent systems on real user computer sessions. The difference is that Adept worked with volunteers, whereas at Meta the subjects were staff employees.

An important nuance: judging by the scale of the protest, the program was launched without separate, explicit notification of each employee. This is a key legal and ethical question: the difference between "consent at hiring" and "informed consent to a specific type of surveillance" is substantial in most jurisdictions.

Why 1,600 people signed the petition

The internal petition, which gathered around 1,600 signatures, is one of the largest known cases of open dissent in Meta's history. The company has no tradition of public internal debate, so a number like this suggests that discontent spread across a significant part of the team rather than staying at the level of isolated complaints.

The employees' key complaints:

• Data collected without the separate, explicit consent of each program participant
• Recording of screen contents, including personal chats, document drafts and confidential work materials
• No clarity on what data is stored, for how long and who has access to it
• Fears that the recordings could be used to assess productivity, not only for AI training
• Violation of the basic sense of privacy, even when working on corporate equipment

Meta employees' monitors display material every day related to future products, partner negotiations, legal documents and internal correspondence. Feeding such data into a training pipeline, even inside the company, raises obvious questions about the limits of corporate privacy.

Meta's response: a pause without details

Meta did not clarify whether the halt is temporary, for example for the duration of an internal audit, or whether the program has been shut down for good. The company also did not say whether work data is being collected by other methods or under other programs.

"We hear the team's concerns and take privacy questions seriously," sources quote

Meta's leadership as saying.

The pause itself is notable. Tech giants rarely back down under pressure from internal petitions, especially on AI initiatives, where competition demands speed. That Meta nonetheless stepped back suggests the scale of discontent exceeded the threshold the company is willing to ignore for development goals.

What this means

The Meta case is one of the first publicly known episodes in which a large corporation collected employee data to train its own AI models and faced mass internal protest. It exposes a key contradiction of the corporate AI boom: the more aggressively a company uses internal data, the more it risks destroying the trust of the people its product depends on.

Regulatory pressure on data collection is already growing, especially in the EU. Now resistance from within is being added to it. It is a signal that other tech corporations considering similar programs are unlikely to ignore.

*Meta is designated an extremist organization and banned in Russia.