Meta Inc. Monitors Employee Activity for AI Training
Meta Inc., led by Mark Zuckerberg, is tracking employee activity across various widely-used websites and applications, including Google, LinkedIn, and Wikipedia. This initiative is part of an effort to gather data for training AI models.
The company has rolled out a new monitoring tool dubbed the Model Capabilities Initiative (MCI), which records employees’ keystrokes and mouse clicks while using company computers. The goal is to gather real-world insights into human-computer interactions, thereby enhancing the development of Meta’s AI agents.
Internal documents reveal that the sites being monitored include not just external platforms but also Meta’s own services. Additionally, third-party tools like Microsoft’s GitHub, Salesforce’s Slack, and Atlassian are on the list. It’s worth noting that this list is evolving, having initially included various AI applications such as OpenAI’s ChatGPT and Anthropic’s Claude.
This data gathering effort aligns with Zuckerberg’s strategy to better compete in the generative AI landscape against companies like OpenAI, Anthropic, and Google. As part of this strategy, Zuckerberg initiated a funding campaign last summer and brought in Alexandr Wang from Scale AI to create a team focused on foundational model development.
However, the announcement of this tracking initiative has raised considerable alarm among employees, with many describing the project as somewhat dystopian in their internal discussions. Concerns have also been voiced regarding potential unintentional exposure of sensitive data, which could include user passwords and personal information related to employees’ health, family, or immigration status.
In response to these concerns, a spokesperson for Meta stated that to develop an effective agent for everyday computer tasks, real examples of usage are necessary. The spokesperson emphasized that safeguards are in place to protect sensitive information and that employee data would not be used for other purposes.
This memo aimed to reassure employees by clarifying that the tracking tool only captures what is visible on an employee’s screen without accessing files or attachments directly. It also stated that any incidental personal information from corporate emails would not be incorporated into AI learning models due to existing safeguards.
Previously reported, Meta has announced that approximately 8,000 employees—about 10 percent of its workforce—will be laid off. This decision was communicated through an internal memo and is set to take effect on May 20. The layoffs are part of a broader strategy to optimize operational efficiency, while the company continues to invest heavily in AI development.
In addition to the staff reductions, Meta has suspended hiring for about 6,000 positions it had previously planned to fill. This dual approach, of cutting existing staff and halting new hires, reflects a comprehensive strategy to control labor costs and streamline operations.
