Meta MCI Program has been paused while the company investigates data security concerns around an internal AI training program.

The program, known as the Model Capability Initiative or MCI, was designed to collect mouse movements, clicks, keystrokes, and other digital activity from U.S.-based employee computers. Meta used this workplace interaction data to train AI models on how people use software and internal systems.

According to Reuters, the Meta MCI Program pause came after internal concerns that sensitive employee data collected through the program may have been accessible to all Meta staffers. Meta says it has no indication so far that employee data was improperly accessed, but the company is putting the initiative on hold while the investigation continues.

Meta confirmed that it will pause the Meta MCI Program, officially called the Model Capability Initiative, while it reviews data security concerns linked to the project.

Company spokesperson Tracy Clayton said Meta had built the program with privacy safeguards and had no indication that employee data was improperly accessed. Still, the company is halting the initiative while the investigation continues.

Reuters reported that the Meta MCI Program pause was still being rolled out as of Monday afternoon, meaning the tool had not yet stopped recording for every affected employee at the time of reporting.

The program was part of Meta’s broader effort to improve AI systems by using real workplace interaction data.

The decision to pause the program followed an internal high-priority security incident report, known at Meta as an SEV.

According to internal documentation reviewed by Reuters, the exposed data included full prompts and transcriptions, private conversations, people and performance data, and DSS sensitivity ratings.

Reuters also reported that one employee involved in the internal discussion said they had accessed personal tax and medical information through a work computer, as had many other employees, while believing the data would be protected and used only for valid business purposes after filtering.

Meta has not said how long the program will remain paused.

The Model Capability Initiative had already drawn attention before the latest pause.

Reuters reported in May that the program was collecting more information than initially described and storing the data in unencrypted form, which raised privacy concerns among employees.

The program sits at the center of a larger tension in the AI industry: companies want more high-quality data to train models, but employees and users are increasingly concerned about how much personal or sensitive information is collected, retained, and reviewed.

AI systems that operate software need to understand how humans use interfaces.

That is why mouse movements, clicks, keystrokes, and workplace activity can be valuable training signals. This kind of data can help AI systems learn how people complete tasks, move through applications, fill forms, interact with internal tools, and recover from mistakes.

But the same data can also be deeply sensitive, especially when collected from employee devices.

This is why internal AI training programs need strong governance before collecting workplace activity.

Meta’s pause highlights a broader challenge for AI companies.

As AI models become more capable, companies increasingly need real-world interaction data to improve them. But collecting that data from employees creates difficult questions around consent, privacy, security, access controls, retention, filtering, and oversight.

The issue is not only whether the data improves AI. It is whether workers clearly understand what is being collected, who can access it, how long it is stored, and whether sensitive personal information can be excluded.

Meta MCI Program pause shows how quickly workplace AI training can become a trust issue.

Even when a company says a program is designed with privacy safeguards, employees may lose confidence if sensitive information appears too broadly accessible. For AI systems built inside large organizations, trust depends not only on model capability but also on data handling.

The investigation will determine whether Meta resumes, changes, or permanently reworks the Model Capability Initiative.

Until then, the pause sends a clear message: AI training programs that monitor employee activity will face heavy scrutiny if data security controls are not strong enough.