Meta Installing Keystroke and Screenshot Tracking Software on Employee Computers to Train AI Models
Meta is installing tracking software on the work computers of employees and contractors based in the US to monitor keystrokes, mouse clicks, movements, and screenshots. The purpose is to use the data for training its AI models. The program, called the Model Capability Initiative, was disclosed in an internal memo shared by a researcher in the company's Superintelligence Labs channel, according to Reuters.
The software will operate on work-related applications and websites such as Gmail, GChat, and Metamate, which is Meta's internal AI assistant for employees. Phones used for work are not included in the tracking.
Meta CTO Andrew Bosworth confirmed that employees using work laptops do not have the option to opt out of the tracking.
What Meta Is Collecting From Employee Work Computers
Meta explains that its tracking system focuses on understanding how employees interact with computers, capturing actions such as selecting options from dropdown menus and using keyboard shortcuts. The company states that this data is meant to help train AI models on tasks that they are not yet able to perform independently.
"Everyone at Meta can contribute to improving our models simply by doing their daily work," the memo noted. Bosworth described the long-term aim as developing AI agents that can carry out tasks while employees oversee and refine their work.
In a statement to CNET, Meta confirmed that the tool collects input data from specific applications to give AI models real examples of human-computer interaction. The company emphasized that the data will not be used in performance evaluations, will not be accessible to managers, and that there are safeguards in place to protect sensitive information.
How Employees and Privacy Experts Are Responding
The internal reaction was described by Business Insider as employees being divided, with some feeling strongly opposed. One employee asked on an internal communications platform how to opt out, and Bosworth responded that there is no option to opt out on work laptops. The staff's reactions included shocked, crying, and angry emoji responses.
Eric Null, director of the Privacy and Data Project at the Center for Democracy & Technology, called the plan one of the most invasive forms of workplace surveillance. He also noted that it could cause real harm to people with disabilities and has the potential to reinforce structural biases if used for AI training.
Why Meta Is Expanding This Tracking Now
The tracking program is being introduced as Meta prepares to lay off roughly 8,000 employees, about 10% of its total 79,000 workforce, starting May 20. The company is investing over $135 billion in artificial intelligence development this year and recently released Muse Spark, its first proprietary AI model created by Superintelligence Labs.
Meta has not provided details on data retention periods, what qualifies as sensitive content under their safeguards, or how interaction data collected will be used in training their models.
Meta has introduced a new internal initiative called the Model Capability Initiative (MCI) that records employee computer interactions—including keystrokes, mouse movements, clicks, and occasional screenshots—to help train its AI models. The goal is to teach AI agents how humans perform everyday computer tasks such as navigating menus, using keyboard shortcuts, and interacting with software applications
If these reports are accurate, this raises serious questions about employee privacy, transparency, and data ethics. While AI companies need large amounts of data to improve their models, monitoring employee keystrokes and screenshots can blur the line between productivity tracking and invasive surveillance. Organizations developing AI should ensure that data collection practices are clearly disclosed, voluntary where possible, and compliant with privacy regulations. Trust and transparency are just as important as technological advancement.
A thought-provoking topic. Organizations must ensure that AI advancement does not come at the expense of employee trust and data security.
Meta employees not only monitored as to what they do on their works computers in order to improve AI models, but then find themselves as collateral damage and no longer wanted!
Always avoid big tech from US. Its bad news.
Meta employees:
We will suck up your data, track and profile you using every instrument & metric possible.
Also Meta employees:
You can’t do this to me! It’s unfair and very wrong. You are violating my privacy!!
The irony appears to be lost on these jerks because as long as it was happening to someone else it was fine but when it happens to them its outrage!
To be clear AI is garbage and these people will likely lose their jobs in the not too distant future so it sucks but my statement still stands.
Very concerning development this highlights the growing tension between AI training needs and employee privacy, especially when real-time monitoring like keystrokes and screenshots is involved
coming to a workplace near you soon, so essentially you are training an ai model for free and have no choice in it. Next item in the list workers will be replaced by said ai
I bet it only works on Windows and maybe Mac, good thing Linux is probably safe from this.
I have no idea why Allwynd thinks the same functions that capture employee screens and keystrokes don’t exist on Linux. They already do. No one is safe.