GoogleLinkedIn and Wikipedia are among hundreds of websites and apps where meta It plans to capture employee keystrokes and mouse clicks as part of a project to train its artificial intelligence models, according to internal messages seen by CNBC.
A new employee tracking tool, called Model Capability Initiative (MCI), allows Meta to observe employees’ actions and collect data on their work computers, Reuters previously reported. informed On Tuesday. List of sites being tracked, including Microsoft’s GitHub, from salesforce dull and AtlassianNot previously reported.
Meta properties like Threads and Manus are also on the list, which is still in flux and basically includes AI apps like OpenAI’s ChatGPT and Anthropic’s Cloud.
The list of third-party sites and services the MCI tool is tracking was widely circulated internally and discussed on chat boards after a member of Meta Superintelligence Labs, or MSL, sent a memo aimed at addressing concerns about worker surveillance and privacy. CNBC saw the memo.
The data gathering project is linked to Meta CEO Mark Zuckerberg’s ambitious effort to catch up in generic AI, where the company has lagged behind OpenAI, Anthropic and others. Google. Zuckerberg began spending last summer trying to bridge the gap, bringing in Alexander Wang of Scale AI to build a team and develop new foundation models.
Earlier this month, Meta unveiled its first major AI model following Wang’s costly appointment. The model, called Muse Spark, marked the debut of the new Muse series developed by MSL, the AI unit that Wang oversees.
Like other tech giants, Meta is working hard on AI agents that can perform various office and coding-related tasks that are usually completed by white-collar workers.
A Meta spokesperson confirmed the project but did not comment on the list of sites being tracked.
“If we are creating agents to help people accomplish everyday tasks using computers, our models need to be real examples of how people do things In fact Use them – like mouse movement, clicking buttons and navigating dropdown menus,” the spokesperson said. ”To help, we are launching an internal tool that will capture this type of input on certain applications to help train our models. “Security measures are in place to protect sensitive content and the data is not used for any other purpose.”
Multiple Meta employees described the data-tracking project as “dystopian” in internal messages seen by CNBC. Others expressed concerns that MCI could widely expose sensitive data, including user passwords, details about new product development and personal information about workers’ immigration status, health or family members.
The MSL employee said in the memo that “to enable our models to use computers” Meta needs a “large and unbiased” data set that reflects how employees work and perform tasks on their corporate devices.
“We needed to capture the on-screen content as the context in which it was being tampered with or interacted with,” the memo said.
Listing “some assurances”, the MSL representative said the new tool would only be able to view employees’ “screen contents” as they see them, and “will not read in files or attachments.”
“Any incidental personal information in your corporate email that may have been screen captured will not be detected by the model due to the above mitigations,” the memo said.
The memo says Meta employees who are still concerned about data-tracking tools can “control what appears on your screen by not doing personal work on your work computer.”
Watch:AI demand metrics are broken and the only thing being anthropogenic is being realistic.
