Meta Abruptly Halts Secret AI Project After Employee Surveillance Data Leak
In a startling turn of events, Meta has been forced to shut down a confidential artificial intelligence training initiative after sensitive employee data, collected through invasive monitoring, was exposed internally. The project, which recorded workers’ keystrokes, mouse movements, and click patterns, was intended to teach AI models how humans interact with computers. However, a major data breach has not only derailed the program but also sparked serious concerns about privacy and trust within the company.
The Secret Project: Model Capability Initiative (MCI)
Launched in April 2026, the Model Capability Initiative (MCI) was developed by Meta’s Super Superintelligence Labs team. The core objective was straightforward yet ambitious: train AI to understand and replicate how people actually use their computers. The system was designed to capture real-world user behavior—everything from choosing options in a dropdown menu to using keyboard shortcuts and navigating between different screens.
To achieve this, Meta began recording the digital activities of its own employees. This included logging every keystroke, tracking every mouse movement, and analyzing click patterns on workstations. According to the company, the data was meant solely for training AI models and was supposed to be protected by robust privacy safeguards.
How the Data Leak Unfolded
The situation escalated dramatically when the collected data inadvertently became accessible across Meta’s broader internal network. According to reports, screenshots viewed by a media outlet revealed highly personal employee chats, performance metrics, and meeting transcripts—all exposed to the entire organization.
Meta internally classified this incident as a SEV 2 event, which is considered a critical threat on a severity scale where 0 is the most severe. The breach effectively shattered the promise that employee data would remain locked down and used exclusively for AI training.
Employee Backlash and Distrust
The leak has ignited significant anger and frustration among Meta’s workforce. From the very beginning, many employees felt uneasy about the mandatory nature of the program. Most staff members were required to participate, leaving them with no choice but to have their every computer action recorded.
One employee posted on an internal company group: “I’m furious. Even though I don’t see evidence of malicious external access, the fact that this data wasn’t fully secure and locked down as promised is deeply disappointing.”
This sentiment reflects a broader erosion of trust. Employees had been assured that the data would be used for no other purpose and would remain completely isolated. The breach has proven those assurances hollow.
What Was Being Recorded?
- Keyboard keystrokes and typing patterns
- Mouse movements and click behaviors
- Workstation activity logs
- Navigation sequences across applications
Meta’s Official Response
A Meta spokesperson released a statement addressing the controversy. “We designed this program with careful privacy safeguards in place. Currently, we have no indication that any employee has misused the leaked data, but we are pausing the initiative until our investigation is complete.”
Despite halting MCI, Meta shows no signs of abandoning its broader data collection ambitions. In a separate internal memo, Chief Technology Officer Andrew Bosworth revealed that the company plans to accelerate internal data gathering under its AI for Work efforts, now rebranded as the Agent Transformation Accelerator (ATA).
The Bigger Picture: Teaching AI to Think Like Humans
Meta’s underlying goal was to develop AI agents capable of assisting people with everyday computer tasks. The company’s logic was straightforward: if AI is to work alongside humans effectively, it must first understand how humans actually behave—how they click, type, scroll, and navigate.
By training models on real employee interactions, Meta hoped to create more intuitive and capable AI systems. The project was seen as a critical step toward building artificial general intelligence that can seamlessly integrate into daily workflows.
Why Was Participation Mandatory?
According to reports, the program was compulsory for the vast majority of Meta employees. This mandatory aspect was a major source of discomfort from the start. Workers worried that their personal activities and behavioral patterns were being captured without genuine consent, blurring the line between professional productivity monitoring and invasive surveillance.
What Comes Next?
While MCI is on hold, the incident has raised pressing questions about corporate surveillance, data ethics, and employee rights in the age of AI. Meta faces the challenge of rebuilding trust while still pursuing its ambitious AI goals. The company’s decision to push forward with the Agent Transformation Accelerator suggests that internal data collection will continue, albeit under a new name and potentially with revised protocols.
For now, the episode serves as a stark reminder that even the most carefully designed privacy safeguards can fail—and when they do, the consequences can ripple through an entire organization.
