Meta tracks employee usage on Google, LinkedIn AI training project
GoogleLinkedIn and Wikipedia are among the hundreds of websites and applications. Meta It plans to capture employees’ keystrokes and mouse clicks as part of a project to train artificial intelligence models, according to internal messages viewed by CNBC.
A new employee monitoring tool called Model Talent Initiative (MCI) allows Meta to observe and collect data from employees’ actions on their work computers; first published by Reuters reported on Tuesday. List of followed sites; Microsoft’s GitHub, Salesforce Loose and Atlassianhas not been reported before.
The list also includes meta features such as Threads and Manus; these are still in flux and originally included AI applications such as OpenAI’s ChatGPT and Anthropic’s Claude.
The list of third-party sites and services that the MCI tool monitors was widely distributed within the company and discussed on chat boards after a member of Meta Super Intelligence Labs, or MSL, posted a memo aimed at addressing employee concerns about surveillance and privacy. CNBC reviewed the memo.
The data collection project ties into Meta CEO Mark Zuckerberg’s ambitious push to catch up with generative AI, where the company has lagged behind OpenAI, Anthropic and Anthropic. Google. To bridge the gap, Zuckerberg went on a spending spree starting last summer, hiring Scale AI’s Alexandr Wang to assemble a team and develop new underlying models.
Earlier this month, Meta unveiled its first major AI model since Wang’s costly hire. The model, called Muse Spark, marked the debut of the new Muse series developed by MSL, the artificial intelligence unit managed by Wang.
Like other tech giants, Meta is pushing heavily for AI agents that can handle a variety of office and coding-related tasks typically performed by white-collar workers.
A Meta spokesperson confirmed the project but did not comment on the list of monitored sites.
“If we are building agents to help people complete daily tasks using computers, our models need real examples of how people work. Actually Use them like mouse movements, clicking buttons, and navigating drop-down menus,” the spokesperson said. “To help, we’re launching an internal tool that will capture these types of inputs from specific apps to help train our models. “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 viewed by CNBC. Others have raised concerns that MCI could widely disclose sensitive data, including user passwords, details about new product development, and personal information about workers’ immigration status, health status, or family members.
To “teach our models how to use computers,” Meta needed a “large and unbiased” dataset that reflects how employees work and perform tasks on their corporate devices, the MSL employee said in the memo.
“We need to capture the content on the screen as the context of what is being manipulated or interacted with,” the memo said.
While listing “a few safeguards,” the MSL representative noted that the new tool will only be able to view employees’ “screen contents” and “will not read files or attachments.”
“Any incidental personal information that may be screen-captured in your corporate email will not be learned by the model due to the above mitigation measures,” the note said.
Meta employees who are still concerned about the data tracking tool can “control what appears on your screen by not doing personal work on your work computer,” the memo said.
WRISTWATCH: AI demand metrics are broken and only Anthropic is behaving realistically.
