ON APRIL 24TH, ACCORDING TO CNBCMeta Through an internal tracking tool called Model Capability Initiative (MCI, Model Capability Plan), keyboard clicks and mouse clicks on hundreds of websites and applications such as Google, LinkedIn, Wikipedia are being recorded as a source of data for training AI models。
An internal memorandum from the members of the Meta SuperIntelligence Laboratory (MSL) indicates that MCI also covers third-party platforms such as GitHub, Slack and Atlassian, and that Meta's own Threads and Manus are also on the list, and that the earlier version of the list included ChatGPT of Openai and Claude of Anthropic。
Meta spokesmen confirmed the project to the media, responding that building smart bodies that help people use computers to perform their daily tasks requires real human use examples, such as "mouse moves, button clicks and drop-down menu operations", and emphasizing that the system has put in place protective measures to protect sensitive content and that data will not be used for other purposes。
MSL further explained in the memo that for the "Culture Model Using Computers", Meta needs a "large and unbiased" data set, that the new tool can only see what is displayed on the employee's screen, "can't read documents or attachments", and suggested that "staff who still have concerns can control what is displayed on the screen by not handling personal affairs on their work computer."。
However, several Meta employees described the project as "anti-Utopia" in their internal sources, worried that the MCI would expose the user code, the details of new product development, and personal information about the employee's immigration status, health status and family members. According to Business Insight, the tracking plan did not provide staff with exit options。