Meta Deploys Employee Surveillance Program to Accelerate AI Agent Training

Meta has launched an internal data-collection project known as the Model Capability Initiative (MCI) to capture the digital behaviors of its workforce. The program tracks keystrokes and mouse clicks on hundreds of websites and applications, including Google, LinkedIn, Wikipedia, GitHub, and Slack, to provide training data for the company's artificial intelligence models. The effort is part of CEO Mark Zuckerberg's aggressive strategy to close the competitive gap with OpenAI, Anthropic, and Google. By observing how white-collar workers navigate software and perform complex tasks, Meta aims to build AI agents capable of automating office and coding workflows. This follows the recent debut of 'Muse Spark,' a model developed by the Meta Superintelligence Labs (MSL) under the leadership of Scale AI's Alexandr Wang. A company spokesperson defended the move, stating that models require real-world examples of button clicks and menu navigation to function effectively. While the company claims safeguards are in place to protect sensitive content, internal memos indicate the tool captures on-screen content to provide context for the interactions being recorded. The program has faced sharp criticism from within the company, with some employees describing the surveillance as 'dystopian.' Concerns center on the potential exposure of sensitive personal information, including passwords and immigration status, as well as proprietary product data. Meta has advised concerned employees to avoid performing personal tasks on corporate devices.