Meta Installs Tracking Software on U.S. Employees' Computers for AI Training
Meta is deploying software to monitor U.S. employees' computer interactions, including mouse movements and keystrokes, to gather data for AI model training. The initiative aims to improve AI agents' ability to perform white-collar tasks. Safeguards are in place to protect sensitive information, according to the company.
marketscreener.comS. employees to capture data such as mouse movements, clicks, keystrokes, and some screenshots, Reuters reported. This data will be used to train the company's AI models. The software will operate on a specific list of work applications and websites.
The effort was outlined in a memo to staff this week from the Meta Superintelligence Labs team, which Reuters reviewed. The memo stated that the data collection would help AI models learn basic computer-use behaviors, such as navigating dropdown menus and using keyboard shortcuts.
Employees were informed they could contribute by performing their regular work tasks.
The initiative supports the development of AI agents capable of handling white-collar tasks independently, amid competition from companies like OpenAI and Anthropic. A Meta spokesperson stated in an email to Fortune that models need real examples of computer use, including mouse movements, button clicks, and menu navigation.
The spokesperson added that the data would not be used for other purposes and that measures are in place to protect sensitive content.
This move occurs as companies seek training data for workplace AI applications. In January, OpenAI was reported to have requested third-party contractors, through training data firm Handshake AI, to submit samples of real work products like PowerPoints and spreadsheets, with instructions to remove confidential material.
Meta acquired a 49% stake in data-labeling firm Scale AI last year for more than $14 billion. Scale AI's former CEO, Alexandr Wang, now leads Meta Superintelligence Labs. Meta CEO Mark Zuckerberg has committed up to $135 billion in capital expenditures for 2026.
The company is also preparing to reduce its workforce by as much as 20%, with initial layoffs scheduled to start in May.
Key Facts
Story Timeline
5 events- Apr 22, 3:03 PM ET
2 new sources added: @alexeheath, BBC News
2 sources@alexeheath · BBC News - This week
Meta disclosed a memo to staff about installing tracking software on U.S. employees' computers for AI training.
1 source@FortuneMagazine - May 2026
Meta's first workforce layoffs are set to begin, as part of a planned 20% reduction.
1 source@FortuneMagazine - January 2026
OpenAI requested contractors to upload real work samples for AI training via Handshake AI.
1 source@FortuneMagazine - Last year
Meta acquired a 49% stake in Scale AI for more than $14 billion.
1 source@FortuneMagazine
Potential Impact
- 01
Workforce reductions may affect up to 20% of Meta employees starting in May.
- 02
Meta's AI models may improve in simulating white-collar tasks, enhancing competitive edge against rivals like OpenAI.
- 03
Employee privacy concerns could arise despite stated safeguards on sensitive data.
- 04
Increased AI spending could strain Meta's financial resources in 2026.
Transparency Panel
Related Stories
cert.europa.euAnthropic Probes Possible Unauthorized Access to Mythos AI Model in Vendor Environment
Anthropic is probing reports of unauthorized access to its Mythos AI model, shared with select companies for vulnerability detection. The incident occurred on the model's release day, involving a Discord group. Separately, Microsoft reduced Game Pass prices but delayed day-one ac…
CnbcTrump Discusses Potential Anthropic AI Deal for Defense Department Amid Iran Ceasefire Deadline
President Donald Trump stated a deal may allow Anthropic's AI models for Department of Defense use following recent White House talks. Separately, he noted restocking by Iran and the U.S. during a ceasefire set to expire tomorrow. Iran prepares new battlefield measures if fightin…
cnet.comOpenAI Introduces New AI Image Feature
OpenAI has released a new feature involving GPT-2 images that blurs the line between artificial intelligence and reality. The feature is described as highly impressive and is available on ChatLLM. Examples of the images can be found on X.