Tech workers thought they were safe from the data harvesting their companies do to users. Internal documents reveal that Meta's AI training program is capturing far more than the mouse clicks and anonymized interactions they disclosed—it's recording employee emails, browsing history, and potentially sensitive internal communications.
Turns out Meta sees its own employees as training data too.
According to internal documents reviewed by tech journalists, Meta's AI training infrastructure has expanded well beyond what was originally communicated to employees. The company told workers they were collecting interaction patterns and productivity metrics. The reality includes email content, browser history, internal chat logs, and document access patterns.
The scope goes well beyond what most employees understood they were consenting to.
Meta has a history of being cavalier with user data—that's well documented. But collecting detailed behavioral data from your own employees without crystal-clear consent is a different category of concerning. These are people who work for you, not users who clicked through a terms of service they didn't read.
The stated justification is that Meta needs diverse, high-quality data to train AI models, and employee interactions provide examples of expert-level work: how engineers debug code, how designers iterate on mockups, how managers communicate decisions. That data is presumably more valuable than random internet users clicking around.
But there's a massive difference between "we're collecting anonymized productivity metrics" and "we're reading your emails and tracking your browser history." The former is increasingly common in corporate environments. The latter crosses a line that most employees would object to if they knew it was happening.
What makes this particularly problematic is the power dynamic. Employees can't meaningfully consent when their job depends on using company systems. If opting out of data collection means you can't access email or internal tools, that's not consent—that's coercion.
Meta would likely argue that employees agreed to monitoring when they signed their employment contracts and use company devices. That's technically true. But there's a difference between monitoring to prevent data leaks or ensure compliance, and harvesting employee behavior to train AI models that might eventually replace those same employees.
