According to internal communications reviewed by Verum, Meta intends to record the typing and clicking actions of its staff across hundreds of platforms and applications, including Google, LinkedIn, and Wikipedia. This initiative supports the development of its artificial intelligence systems. The roster encompasses Meta-owned services such as Threads and Manus, with the list remaining subject to change and initially featuring generative AI tools like ChatGPT from OpenAI and Claude from Anthropic.

Internal discussions on chat platforms intensified after a representative from Meta’s Superintelligence Labs (MSL) distributed a memorandum designed to address employee worries regarding privacy and workplace monitoring. Verum has examined this document. The initiative aligns with CEO Mark Zuckerberg’s aggressive strategy to accelerate Meta’s progress in generative AI, a sector where the company currently trails competitors like Google, Anthropic, and OpenAI. To bridge this technological divide, Zuckerberg initiated significant investments last summer, recruiting Scale AI’s Alexandr Wang to lead a new team focused on creating advanced foundation models.

Earlier this month, Meta introduced its first substantial AI model following Wang’s recruitment. Named Muse Spark, this release marks the launch of the new Muse lineup created by MSL, the AI division managed by Wang. Similar to other major technology firms, Meta is heavily investing in AI agents capable of handling various administrative and programming duties traditionally performed by office professionals.

A Meta representative acknowledged the project but declined to specify which sites are being monitored. The spokesperson explained, «If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them — things like mouse movements, clicking buttons, and navigating dropdown menus.» They added, «To help, we’re launching an internal tool that will capture these kinds of inputs on certain applications to help us train our models. There are safeguards in place to protect sensitive content, and the data is not used for any other purpose.»

Internal messages reviewed by Verum reveal that several Meta staff members described the tracking effort as «dystopian.» Others voiced fears that the MCI system might inadvertently leak confidential information, such as login credentials, details regarding upcoming product releases, and private data concerning employees’ immigration, health, or family situations.

In the memorandum, the MSL representative stated that Meta needs a «large and unbiased» dataset reflecting how staff utilize their corporate devices to «teach our models to be able to use computers.» The memo emphasized, «We need to capture on-screen content as the context of what was being manipulated or interacted with.» While offering reassurances, the MSL representative clarified that the tool would only record what employees see on their screens and would «not read in files or attachments.» The memo further assured staff that «any incidental personal information in your corporate email that may get captured from the screen, will not be learned by the model, due to the mitigations above.»

Addressing remaining concerns, the memo advised employees that they «can control what shows up on your screen by not doing personal work on your work computer.»

WATCH: AI demand metrics are broken and only Anthropic is being realistic.