The news: Meta will start using US-based employees’ computer activity to train its AI models. It’s installing tracking software that will capture mouse movements, clicks, and ​keystrokes to help the company build AI agents, per Reuters.

“If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them,” a Meta spokesperson said, per TechCrunch.

There is no opt-out option, per Business Insider, though the company stated that safeguards are in place to protect sensitive content. Some employees had immediate negative reactions to the news, including stating discomfort and using crying and angry emojis on Meta’s internal workplace communications site.

Zooming out: Meta is working hard to keep up in the AI race, including by heavily spending on talent and infrastructure, but it has faced challenges.

• Prior to the release of its Muse Spark model this month, Llama 4 received a lukewarm welcome a year ago.
• Meta reorganized its specialized Meta Superintelligence unit last year and laid off 4,300 employees, per Layoffs.fyi.
• The company has already laid off 1,700 in 2026 and plans to cut about 10% more, or 8,000 employees, in May, per Reuters. This AI training effort could be part of a plan to replace employees with AI agents by training them on real humans’ workflows.

Meta has determined that synthetic and scraped data isn’t enough for building truly useful agentic AI and that it needs granular, real-world behavioral data to compete. Other Big Tech firms could follow suit—especially those building computer-use agents, like Google and Microsoft—by turning employee workflows into free training pipelines.

The challenge: Even with safeguards around employee data, keystroke-level monitoring is different from typical productivity tracking and is likely to be seen as an invasion of privacy. It could have an impact on morale, retention, and internal culture, especially in a competitive talent market.

This news could also affect user trust around Meta’s platforms considering the company already uses some consumer data—including publicly shared Facebook, Instagram, and Threads posts—to train its models.

Implications for the industry: Enterprise environments could turn into data sources for AI companies as licensing deals become more expensive and potential consequences—both legal and reputational—grow around data scraping.

• Employees are becoming direct training sources, raising concerns around consent and compensation and risking backlash or even worker departures.
• This kind of tracking could attract regulatory or legal scrutiny—which may be why it’s only being implemented in the US and not the EU—potentially limiting how human behavioral data can be used in AI training.

This content is part of EMARKETER’s subscription Briefings, where we pair daily updates with data and analysis from forecasts and research reports. Our Briefings prepare you to start your day informed, to provide critical insights in an important meeting, and to understand the context of what’s happening in your industry. Non-clients can click here to get a demo of our full platform and coverage.