Meta’s AI Tracking Tool Raises New Workplace Privacy Questions

Meta’s internal AI training project is raising a larger question for the future of work: how much employee behavior should companies be allowed to record when building AI agents?

The issue centers on Meta’s “Model Capability Initiative,” a workplace data collection tool designed to capture how employees use computers so the company can train AI systems to perform everyday software tasks. According to Reuters, the tool tracks actions such as mouse movements, clicks and navigation across more than 200 apps and websites.

Meta has said the program is aimed at U.S. employees and is intended to help build AI agents that can complete computer-based tasks. But Reuters reported that internal documentation and employee complaints suggest the tool may also capture data involving people outside the United States when they communicate with U.S.-based colleagues. That creates a potential privacy problem in Europe, where data protection rules are stricter.

The concern is not just that an employer can monitor work devices. Many companies already use security tools, access logs and compliance systems. The difference is the level of detail and the purpose of the collection. Meta’s tool is not only watching for security threats or policy violations. It is gathering examples of how people work so AI systems can learn to perform similar tasks.

That makes the debate more complex. AI agents need high-quality examples of real computer use to become useful. A system that can book meetings, update documents, navigate software tools or complete repetitive office tasks must learn from human workflows. But collecting those workflows can also expose messages, documents, code changes, clipboard content and patterns of behavior that workers may not expect to become AI training data.

Reuters reported that some employees complained the tool was consuming large amounts of data and capturing extensive behavioral information. Earlier in May, Reuters also reported that Meta employees distributed flyers at U.S. offices protesting the tracking software, with some workers seeing it as part of a broader shift toward AI-driven workplace automation.

For employees, the core question is consent and control. If a company owns the device and the work account, it may have broad monitoring rights in some jurisdictions. But privacy expectations change when data is repurposed from workplace activity into AI training material. A chat message, code review or document edit may have been created for work, but that does not automatically mean every detail should become training data for an AI system.

In Europe, the issue may be even sharper. The General Data Protection Regulation requires companies to have a lawful basis for processing personal data and to clearly explain how data is used. If communications involving EU-based workers are captured through a U.S. colleague’s device, regulators may ask whether those individuals received enough notice or had meaningful control over the processing.

This is why the Meta case matters beyond Meta. Many companies are racing to build or deploy AI agents that can automate office work. To make those agents effective, they may want examples of real workflows from employees. But the more realistic the data becomes, the more likely it is to include sensitive or personal information.

There is also a trust problem. Workers may worry that the data used to train productivity tools could later be used to evaluate performance, restructure teams or replace roles. Even if a company says the data is anonymized or not used for performance reviews, employees may still question how long that promise will last and who can audit it.

Meta has argued that real examples of computer use are needed to build agents that can help people complete tasks. That is a legitimate technical challenge. AI systems often struggle with the messy details of workplace software: menus, forms, approvals, file naming, internal tools and multi-step processes.

But solving that challenge through broad employee tracking creates a policy problem. Companies may need clearer rules on what can be collected, what must be excluded, how long data is retained, whether workers can opt out and whether third parties or regulators can verify safeguards.

For ordinary workers outside Meta, the story is a preview of a larger workplace shift. AI is not only changing what tools people use. It is changing what parts of their work become data.

The next phase of AI in the office may depend less on flashy chatbots and more on quiet observation: clicks, shortcuts, emails, files and repeated routines. That could make software more helpful, but it could also make workplace surveillance harder to avoid.

The Meta tracking debate shows that AI agents will not only be judged by what they can do. They will also be judged by how the data used to train them was collected.