Meta: Employee Monitoring by AI Sparks Debate

Meta Adjusts Its Controversial MCI Program

Meta has recently adjusted its MCI (Model Capability Initiative) following internal criticism. These adjustments include enhanced privacy protections and the option for certain employees to request an exemption. A new feature also allows users to suspend tracking of activity for 30 minutes. These changes were prompted by employee concerns regarding the software's impact on battery autonomy and the potential presence of personal data on work devices. Employees have also called for better control over the information collected on-screen.

Recap of the MCI Program Launch

Last April, Meta launched the MCI program, an initiative aimed at turning its employees into study subjects to train its AI. The program, known as the Model Capability Initiative, aims to observe how users interact with their computers by monitoring mouse movements, clicks, and menu navigation. The goal is to develop AI agents capable of autonomously performing everyday computing tasks. According to information shared with employees, MCI monitors over 200 applications and websites.

Meta initially assured that the system was limited to employees based in the United States and that it included mechanisms to protect sensitive data. However, not everything went as planned, and some employees expressed their opposition to these practices.

Employee Issues Encountered

A few weeks after the rollout of MCI, employees began reporting issues. According to internal documents reviewed by Reuters, some claim that the tool generates a significant volume of data, causing spikes in home internet usage. These spikes are so substantial that they can deplete a monthly data plan in just a few days. Meta also acknowledged that the system could record the content of emails and direct messages sent to American employees, even when the senders are located in other countries.

In a statement to Reuters, Dave Arnold, a spokesperson for Meta, clarified that MCI is only installed on the devices of American employees. He emphasized that the primary goal of the program is to analyze interactions with computers, not the content displayed on the screen. According to him, employees located outside the United States are informed of these practices.

Additional Revelations and Concerns

In an internal post, an employee shared the results of an analysis of MCI's technical logs. This study suggests that the tool may have access to numerous additional pieces of information due to its integration with existing security software on the company's devices. The analysis notably claims that the system could record code changes, computer sleep cycles, web addresses visited, and content copied and pasted into the clipboard. Some of this data may have even been stored in a less secure manner than anticipated.

The author of the message believed that such a level of data collection would allow for an accurate reconstruction of an employee's work methods. According to him, the goal would not simply be to create an AI that clicks in place of a user, but to develop a sufficiently advanced system to understand each step of a work process and execute it autonomously. The message subsequently disappeared from internal discussion spaces, according to employees interviewed by Reuters. Meta disputes these claims, and Dave Arnold described the analysis as “fundamentally inaccurate,” though he declined to comment in detail on the points raised or the circumstances surrounding the message's disappearance.