Meta is implementing tracking software on the computers of its employees in the US.

      According to an internal memo reviewed by Reuters, Meta has announced that it will implement new tracking software on the work computers of employees based in the U.S. This software will monitor mouse movements, clicks, keystrokes, and occasional screenshots, with the gathered information being utilized to train the company's AI models. Critics argue that this initiative is merely a form of workplace surveillance disguised in another way.

      The tool, known as the Model Capability Initiative, was revealed to the staff in a memo circulated through a channel associated with the Meta Superintelligence Labs team. As per the memo, which Reuters accessed, the software will operate on a specified list of work applications and websites. Employees were informed that they can contribute by simply performing their daily tasks.

      Meta claims that to develop agents that can operate software on behalf of users, it is essential to collect training data from actual human computer interactions, particularly the minute behaviors, such as selecting items from dropdown menus and using keyboard shortcuts, that AI models find difficult to replicate with general web data.

      A spokesperson from Meta stated, “In building agents to assist with everyday computer tasks, our models require authentic illustrations of how individuals utilize them—elements like mouse movements, button clicks, and interacting with dropdown menus.” The company mentioned that measures are in place to safeguard sensitive information and that the data collected will solely be used for model training.

      The memo positioned the initiative as part of Meta’s "AI for Work" program, which has been rebranded as the Agent Transformation Accelerator. The division responsible for this initiative, Superintelligence Labs, is now led by Alexandr Wang, the former CEO of Scale AI, a data-labeling company in which Meta acquired a 49% stake for over $14 billion last year.

      This connection is noteworthy because Scale AI specializes in generating training data on a large scale, and Wang's appointment indicates that Meta views data collection as a pivotal strategic asset rather than merely a procurement activity.

      This situation raises significant questions about the labor dynamics between AI firms and their workforce. For Meta employees, the relationship is particularly direct: they are, in essence, being asked to produce the training data that will help AI agents mimic their own computer usage.

      Meta is not the first organization to leverage internal workflows for AI training purposes; earlier in January, OpenAI was reported to have requested contractors to provide samples of their actual work. However, the combination of keylogging and screenshot capturing represents a more comprehensive and automated strategy.

      Despite Meta's assurances that the data collected will not be used for monitoring employee performance, public responses have been skeptical, although there is no evidence indicating such practices currently exist. The tracking software will function only on the specified list of applications and websites, and Meta has not revealed which apps are included in this list. Currently, the data collection seems restricted to U.S. employees; compliance with EU and UK data protection laws would necessitate clear legal grounds and possibly explicit consent for similar initiatives in those regions.