Meta Plans to Monitor Employee Keystrokes and Screen Activity to Enhance AI Training Efforts
Meta plans to gather employee mouse movements, keystrokes, and screen data to enhance AI models through a new internal initiative.
Meta plans to start monitoring employees’ mouse movements, clicks, and keystrokes on work devices as part of a significant initiative to enhance its artificial intelligence systems, as indicated by internal memos reviewed by Reuters.
The newly introduced tool, known as the Model Capability Initiative (MCI), is designed to oversee activities across work-related applications and websites, while also periodically taking screenshots of employees’ screens. The data will serve to train AI models to more accurately replicate human interactions with computers.
One internal memo indicates that the objective is to enhance AI capabilities in areas like navigating dropdown menus and utilizing keyboard shortcuts.
“This is where all Meta employees can contribute to improving our models just by performing their daily tasks,” the memo stated.
This initiative is a key component of Meta’s extensive plan to embed AI thoroughly within its operations and create autonomous agents that can perform various workplace tasks.
In a separate memo, Meta Chief Technology Officer Andrew Bosworth stated that the company was speeding up internal data collection as part of its “AI for Work” initiative, which has now been renamed the Agent Transformation Accelerator (ATA).
“The vision we are striving for is one in which our agents take on the primary responsibilities, while our role is to guide, assess, and assist them in their development,” Bosworth stated.
“The goal, he noted, was for agents to “instantly recognize where we believed intervention was necessary so they could improve in the future.”
According to Meta spokesperson Andy Stone, the data collected through MCI will be exclusively utilized for the purpose of training AI models and will not be employed in employee performance evaluations.
“To create agents that assist individuals in completing daily tasks on computers, it is essential for our models to have authentic examples of actual user interactions — such as mouse movements, button clicks, and navigating dropdown menus,” Stone stated.
The initiative arises as Meta reorganizes its workforce with a focus on AI. The company intends to reduce approximately 10% of its global workforce beginning May 20, with additional cuts anticipated later this year.
Other significant technology companies have implemented comparable measures. In recent months, Amazon has decreased its corporate workforce by approximately 30,000 positions, while fintech company Block has laid off nearly half of its employees.
Meta has internally motivated employees to increasingly utilize AI tools for coding and various tasks and has launched a new general-purpose position called “AI builder.” Recently, the company has also established an Applied AI engineering team to enhance the coding capabilities of its systems.
Nonetheless, the plan has raised concerns about workplace surveillance. Ifeoma Ajunwa observed that although monitoring tools have traditionally been employed to identify misconduct, keystroke logging signifies a more profound level of scrutiny, raising concerns about employee privacy and trust in the workplace.
“On the US side, federally, there are no restrictions on worker surveillance,” she stated.
Valerio De Stefano noted that such practices would probably encounter legal obstacles in Europe, where more stringent data protection regulations are in effect, particularly due to the General Data Protection Regulation (GDPR) that emphasizes individual privacy rights and limits the extent of surveillance that employers can conduct.
“He noted that, more generally, the awareness of employer surveillance alters the power dynamics in the workplace, favoring the employer.”
