Meta has halted its Model Capability Initiative (MCI), an internal program that tracks employee mouse movements, clicks and keystrokes to train AI models.
The company shut down the program on June 22, 2026, after documents reviewed by Reuters showed that sensitive employee data collected through MCI was inadvertently accessible to all Meta employees.
It has not been announced when the program will restart or whether it will return to its current form. The data exposure was not caused by an external breach, but rather resulted from a permissions misconfiguration that allowed all Meta employees to access data collected by the program.
What does the Model Capability Initiative do and what employee data was exposed
Model Capacity Initiative was launched in April 2026. It collects various data points from US-based Meta employees, including mouse movements, clicks, keystrokes, and the occasional screenshot.
This collected data is used to train Meta’s AI models, which aims to help AI systems better understand how human workers perform tasks. However, according to Reuters, the initiative resulted in the collection of a large pool of detailed employee behavior data.
The data reportedly exposed included private conversations, performance data and transcriptions. Typically, such content would be limited to HR or a small group of managers. However, due to a permissions error, the data was accessible to Meta’s entire workforce.
Meta spokesperson Tracy Clayton said, “We have carefully designed this program with privacy safeguards in place. While we have no evidence at this time that any data was improperly accessed by Meta employees, we are pausing the program to investigate.”
The company also refused to say how long this ban would last.
Internal tensions, legal risks and what comes next for Meta’s surveillance program
MCI has been a controversial topic within the meta since its inception. Employees raised concerns about being monitored by software designed to analyze their work habits, especially since the program was launched ahead of a series of major layoffs, including an announcement of 8,000 job cuts in May 2026.
To help alleviate some of these concerns, Meta added a pause feature that allows employees to disable tracking for up to 30 minutes at a time. This addition revealed how widespread surveillance has become.
Logging keystrokes and taking screenshots of identifiable employees can pose serious legal risks under the EU’s General Data Protection Regulation.
GDPR standards are strict when it comes to processing personal data and workplace consent is generally considered unreliable given the power imbalance between employers and employees.
The scenario where sensitive employee records within a company become widely accessible represents the very failure that the GDPR seeks to prevent.
Although the program is currently limited to United States employees, Meta’s long-term plans for similar features will face considerable regulatory challenges in the European Union.
Meta said it would investigate the exposure. The company did not provide any timeline for the investigation, a resumption date for the program, or details about any changes that would need to be made before MCI returns.
The incident raises broader questions about employee oversight for AI training. More companies are using tools that monitor work patterns to help train assistants and agents. Such programs require strict internal access controls that match the sensitivity of the data involved.
At present MCI is offline. Meta has not said whether the program will return to its current form, be redesigned, or be discontinued.
Thanks for being a Ghax reader. The post Meta halts employee mouse-tracking AI training program after internal data exposure appeared first on gHacks.
