To The Moon

Meta's plan to use its employees' daily activities to train its AI systems has met with strong internal resistance, forcing the company to implement changes.

The Information reports that Meta has recently made several restrictive adjustments to its employee monitoring tool, the "Model Capability Initiative" (MCI), due to large-scale employee opposition. The adjustments include allowing employees to pause monitoring for 30-minute intervals, expanding the categories of staff eligible for exemptions, and changing the data collection method from recording precise keystrokes to generating "activity summaries."

However, these changes represent a modification rather than a termination of the program. The MCI system remains operational, and Meta's broader AI training strategy remains unchanged.

What the Tool Tracks

The MCI was launched in April and was introduced internally by an AI research scientist from Meta's model-building team, Meta SuperIntelligence Labs.

Its operation is straightforward: tracking software is deployed on employee work computers to collect mouse movements, clicks, keyboard operations in real-time, and periodically capture screen content. This covers work-related applications and websites.

Meta's stated rationale is that current AI models have shortcomings in simulating the nuances of human-computer interactions—such as selecting options from drop-down menus or using keyboard shortcuts—which are precisely the tasks human employees perform daily.

CEO Mark Zuckerberg explained the logic during an all-hands meeting in April, stating that AI models are currently in a phase where they "learn by watching smart people do things," making it crucial for systems to observe these actions. He further justified using Meta's own employees over outsourced data labeling firms, suggesting the average intelligence level within the company is significantly higher.

A Meta spokesperson confirmed that data collected by MCI would be used as input for AI training and asserted it would not be used for employee performance evaluations.

How Employees Fought Back

Employee resistance was direct.

Some developed a habit of ignoring the authorization pop-ups, while others found ways to disable the software through device settings. Others reported that the MCI installation noticeably slowed down their devices.

The resistance quickly escalated from individual actions to collective efforts. Last month, some Meta employees posted flyers in several U.S. offices, urging colleagues to sign an online petition demanding that Zuckerberg and management halt the data collection. Reuters reported the petition garnered over 1,500 employee signatures.

Petition organizers also released a statement calling on the company to "treat employees and contractors with more dignity."

Meta's Concessions and Unchanged Core

Facing this pressure, Meta announced several adjustments in an internal memo.

A new pause function allows employees to suspend MCI for up to 30 minutes for "viewing personal matters," after which it automatically resumes.

The exemption criteria have been broadened. Previously limited to roles handling confidential information, exemptions now extend to employees dealing with sensitive content, remote workers concerned about bandwidth costs, and field staff who struggle to keep laptops charged.

Privacy protections have been ostensibly enhanced. The data collection method has shifted from recording precise keystrokes to generating "activity summaries." The company also stated that due to strict internal access controls, only "a small number of engineers" can access the raw data collected by MCI.

However, the core framework remains intact: MCI is still running, employees are still a source of AI training data, and the automatic resumption after 30 minutes means it is not an option that can be permanently switched off.

The Bigger Picture: Employees in the AI Pipeline

The MCI is just one part of Meta's larger strategy.

According to Reuters, Chief Technology Officer Andrew Bosworth described the company's direction in another internal memo, envisioning a future where "agents do most of the work, and our role is to direct, review, and help them improve." This strategic project has been named the "Agent Transformation Accelerator" (ATA).

Simultaneously, Meta has internally begun requiring employees to use AI agents for tasks like coding, even if it reduces short-term efficiency. The company is also blurring functional boundaries between some roles, promoting a generic position called "AI builder."
<
In other words, employees are both the data source for training AI and the future subjects of potential AI replacement.

The Legal Landscape

In the United States, such monitoring currently faces no federal restrictions. A Yale Law School professor noted that state laws at most require employers to provide broad notice of monitoring. She also pointed out that keystroke logging subjects white-collar employees to a level of real-time surveillance previously limited to delivery drivers and gig economy workers.

The situation is different in Europe. A law professor specializing in technology and comparative labor law stated that such monitoring would likely be deemed illegal there. Italy explicitly prohibits electronic monitoring for tracking employee productivity, and German courts have ruled that keystroke logging is only permissible under special circumstances, such as suspicion of serious criminal activity. The professor also suggested the practice might violate the EU's General Data Protection Regulation (GDPR).

He further observed that the rise in employer monitoring awareness is shifting the broader workplace power dynamic, tilting the balance further in favor of employers.