Meta has paused its employee tracking software, the Model Capability Initiative (MCI), after a data leak exposed sensitive employee activity logs. The tool, designed to train AI models, inadvertently left data like keystrokes and screenshots accessible within the company, prompting an investigation and internal criticism.
Meta faces internal chaos regarding its AI strategy, employee tracking, and layoffs. Employees express dissatisfaction with AI reorgs, data collection, and severance packages, while scammers exploit hotel reservation data and the Pentagon struggles with troop phone tracking.
Meta is making minor concessions to its controversial Model Capability Initiative (MCI) for AI training, allowing employees to pause tracking for 30 minutes and offering limited opt-outs. The majority of employees will still be tracked, despite protests and concerns about surveillance.
Meta is deploying a new internal tool called Model Capability Initiative (MCI) to capture employee keystrokes, mouse clicks and screen content across hundreds of websites and apps including Google, LinkedIn, Wikipedia, GitHub and Slack for AI training purposes. The project, confirmed by a Meta spokesperson, aims to collect real examples of how people use computers to train AI agents. Multiple employees internally characterized the data collection as 'dystopian,' expressing concerns about exposure of sensitive data including passwords, product development details and personal information. The initiative is tied to CEO Mark Zuckerberg's effort to close the AI gap with OpenAI and Anthropic, including the recent hiring of Scale AI's Alexandr Wang.
