Meta AI Project: Spying on employees backfires on Meta; secret AI project shut down—here are the details..

Meta AI Training Program: Meta launched an internal AI training program called the Model Capability Initiative (MCI) in April 2026. It was developed by the company's Meta Super Superintelligence Labs team.

The program aimed to teach AI models how humans use computers. Employees' actual usage patterns—such as selecting items from dropdown menus, using keyboard shortcuts, or navigating across different screens—were recorded to explain these activities to the models.

What employee activities were being recorded?
According to reports, the system recorded activities such as keyboard keystrokes, mouse movements, clicking patterns, and general workstation activity.

Meta stated that this data would be used solely to train AI models and that adequate privacy safeguards were in place.

How did the data leak occur?
According to a media report, the situation escalated when the internal data became accessible across the company's broader network.
Screenshots viewed by Business Insider revealed that highly personal employee chats, performance data, and meeting transcripts were exposed to the entire company.

SEV 2 Category Incident: Given the severity of this massive data leak, Meta has internally classified it as a major threat under the SEV 2 category (on a scale of 0 to 5, where 0 represents the most critical level).

Growing employee dissatisfaction
Employees questioned the company's data security measures following such a large-scale leak. In fact, employees were already uncomfortable when the program was introduced, as participation was made mandatory for most staff. Meta had initially promised that the data would not be used for any other purpose and would be kept completely secure (locked down). However, that did not happen, leading to continued resentment among employees. One employee even wrote in the company's internal group, "I am very angry. Even though I see no evidence of external or malicious access, it is extremely disappointing that this data was not fully secured and locked down as originally promised."

What did Meta say?
A Meta spokesperson also stated the matter. They said, "We designed this program very carefully with privacy safeguards in place. Currently, we have found no indication that any Meta employee has misused the leaked data, but we are pausing the program pending a full investigation."

However, despite this pause, Meta is not backing away from its plans to gather data for AI. In a separate memo, Meta CTO Andrew Bosworth revealed that the company would further accelerate internal data collection as part of its "AI for Work" efforts—now rebranded as the "Agent Transformation Accelerator" (ATA).

Attempting to make AI models human-like
Meta envisioned developing AI agents in the future that could assist people with their everyday computer activities.
According to the company, if AI is to be taught to function like a human, it is essential to understand how people actually interact with computers.
This is why AI models were being trained based on the actual activities of employees.

Why was it mandatory for employees?
According to reports, the program was mandatory for most of the company's employees. Consequently, many employees expressed discomfort with the initiative from the very beginning. They were concerned that their personal life activities and behaviors were being recorded.

Disclaimer: This content has been sourced and edited from Amar Ujala. While we have made modifications for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.