Meta Spies on Its American Employees to Train Its AI

Meta Monitors Its American Employees to Train Its AI

Internal memos from Meta have revealed the implementation of monitoring software on the computers of its employees based in the United States. This software records clicks, keystrokes, and mouse movements to feed future artificial intelligence agents of the company.

Meta has already been involved in several lawsuits in Europe regarding the use of its users' data for AI training. The internal memo, discovered by Reuters, mentions a tool called the Model Capability Initiative, which operates on a defined list of professional applications and websites, with occasional screenshots to contextualize behavioral data.

Andy Stone, a spokesperson for Meta, stated that this data would not be used to evaluate employee performance but to provide real-world examples necessary for building agents capable of assisting with daily tasks.

An Unintentional Contribution to AI Replacement

The memo, shared within the Meta Superintelligence Labs team, informs employees that they can contribute to the improvement of the company's models simply by performing their daily work. However, Meta plans to assign this same work to autonomous agents. The company is considering laying off approximately 8,000 employees on May 20, which represents 10% of its global workforce, with further reductions expected in the second half of 2026.

These job cuts are accompanied by an increase in AI investments, with capital expenditures estimated between $115 and $135 billion for 2026, roughly double what Meta had invested the previous year. Engineers from across the company have already been transferred to an applied AI division tasked with building agents capable of executing code and complex tasks autonomously.

An Initiative Limited to the United States

The monitoring program only affects Meta's American employees due to legal constraints in Europe. In Italy, electronic monitoring of employees for productivity purposes is explicitly illegal. In Germany, recording keystrokes is only permissible under exceptional circumstances, such as suspicion of a serious criminal offense. This practice would constitute a violation of the GDPR.

Meta has already faced several legal conflicts in Europe regarding the use of its users' data for AI training. The European organization noyb had denounced a blatant violation of the GDPR when Meta attempted to exploit the posts of its European users to train its models, an initiative that Meta ultimately suspended under regulatory pressure.

U.S. labor law does not impose such restrictions on workplace monitoring, unlike several European legislations. OpenAI, for example, had solicited contractors to submit real professional documents, asking them to remove confidential information before sending. Meta is now adding its employees' actions to the training of an AI that could, in theory, replace them. In Europe, the GDPR still protects us from these practices, but for how long?