Meta Will Track Employees’ Keystrokes, Clicks and Mousing to Train AI

Meta will track its employees’ keystrokes, clicks and mouse movements — and even capture screenshots of what’s on their computer screens — to help train the company’s AI models. That’s according to a Reuters report on Tuesday, citing an internal memo sent to workers.

According to the memo, Meta will install a new software program called the Model Capability Initiative on the computers of US-based employees and contractors. The tracking software will operate on work-related apps and websites and is part of Meta’s plan to build AI agents that can do tasks autonomously.

The announcement, published in its entirety by Business Insider, said that monitored apps and URLs would include Gmail, GChat and Metamate, an employee AI assistant. Workers’ phones would not be included in the tracking.

Business Insider reported that Meta employees were “up in arms” about the plan to use tracking software.

On an internal communications website seen by the news outlet, one employee wrote, “This makes me super uncomfortable. How do we opt out?”

Meta CTO Andrew Bosworth responded, “There is no way to opt out on your work laptop,” prompting staff to react with shocked, crying and angry emoji, according to Business Insider.

As it invests in AI development — more than $135 billion this year — Meta continues to reduce headcount. The company plans to lay off about 8,000 employees, 10% of its workforce of 79,000, starting May 20. The company reportedly has cut 25,000 jobs since 2022.

Meta’s AI surveillance

Meta wants to train its AI on tasks it cannot yet replicate, focusing on how people actually use their computers. This includes such actions as selecting options from dropdown menus and using keyboard shortcuts.

“This is where all Meta employees can help our models get better simply by doing their daily work,” the memo said.

Reuters said the memo was posted by an unidentified AI research scientist on Tuesday in a channel for the company’s Superintelligence Labs team.

According to Reuters, Bosworth told employees that the long-term vision was for AI agents to “do the work” while employees direct them and help them improve. He did not specifically say how the agents would be trained with the data, but did say that Meta would rigorously gather data “for all the types of interactions we have as we go about our work.”

Eric Null, director of the Privacy and Data Project at the digital rights organization Center for Democracy & Technology, said Meta’s plan to track employee computer interactions is one of the most “invasive” forms of workplace surveillance.

“That invasiveness underscores the need for clear privacy protections and AI guardrails,” Null told CNET. “This type of surveillance can cause real harm to people with disabilities, and workers in general chafe at this kind of tracking. Using this data for AI training in particular has the potential to replicate structural biases.”

In a statement given to CNET, a Meta spokesperson said that tracking employees is intended to give AI models “real examples” of how people interact with their computers.

“To help, we’re launching an internal tool that will capture these kinds of inputs on certain applications to help us train our models,” the spokesperson said. “There are safeguards in place to protect sensitive content, and the data is not used for any other purpose.”

Meta said it would not use the collected data in performance reviews and that managers would not be able to see it.

Business Insider cited an unnamed source saying that, when hired, employees are told their work devices can be monitored by Meta.

Bad optics

Bill Howe, associate professor at the University of Washington’s Information School, said that, despite the bad optics, Meta assumes it can extract a lot of value from harvesting data from employees’ computer interactions.

“Employees everywhere are helping to train the systems that will take their jobs,” Howe told CNET.

Meta appeared to raise its standing in the AI race earlier this month when it unveiled Muse Spark, the first AI model developed by its Superintelligence Labs. This week, Meta also broke ground on its planned data center in Tulsa, Oklahoma, which, when completed, will be the company’s 28th in the US.

Howe said that the federal government should pass new tax laws to “counteract the runaway inequity that is emerging,” as multibillion-dollar tech companies continue boosting profits for investors while cutting down on staff.

“As Meta is demonstrating, companies ultimately are not incentivized to care about workers, so we need solutions at the federal level,” Howe said.