NEW YORK, April 21 - Meta is installing new monitoring software on the computers of employees based in the United States to record their mouse movements, clicks and keystrokes, with occasional snapshots of on-screen content, the company informed staff in internal memos examined by the company. The tracking program, named Model Capability Initiative or MCI, will operate on work-related apps and websites and is intended to generate training data for internal artificial intelligence models designed to perform work tasks autonomously.
According to a memo circulated within one of Meta's model-building teams and posted by a staff AI research scientist, MCI's collected signals are expected to help models learn interactions that are currently difficult to reproduce, such as selecting items from dropdown menus or employing keyboard shortcuts. The memo framed participation as part of routine work: "This is where all Meta employees can help our models get better simply by doing their daily work," it said.
The tracking initiative comes amid a wider push by Meta to bake AI throughout its operations and to reorganize roles around AI capabilities. In a separate note to staff, Chief Technology Officer Andrew Bosworth described a shift in which AI agents would carry out much of the work and human employees would concentrate on directing, reviewing and improving those agents. "The vision we are building towards is one where our agents primarily do the work and our role is to direct, review and help them improve," Bosworth wrote. He added that a goal is for agents to be able to "automatically see where we felt the need to intervene so they can be better next time."
Bosworth said Meta intends to be methodical about the process, stating the company would be "rigorous" in assembling data and evaluations for the varieties of interactions employees have while performing their jobs. While he did not detail the precise training regimen for agents, a company spokesperson acknowledged that the MCI data would be among the inputs used to enhance model capabilities.
Meta spokesperson Andy Stone told staff that information gathered through MCI would be used for model training and not for assessing employee performance or other purposes. Stone also said safeguards are in place to shield "sensitive content," but did not specify which types of data would be excluded from collection. "If we’re building agents to help people complete everyday tasks using computers, our models need real examples of how people actually use them - things like mouse movements, clicking buttons, and navigating dropdown menus," he said.
The company has been encouraging employees to leverage AI agents for activities such as coding and other operational tasks, even if those tools slow short-term productivity, as part of a campaign to accelerate adoption. Meta has also been consolidating some job distinctions into a new, more general job title called "AI builder." Most recently, the company formed a new Applied AI (AAI) engineering team focused on boosting the coding capabilities of Meta's AI models and using those models to craft agents capable of performing the bulk of work needed to build, test and ship products and infrastructure.
Internal movement has already begun. Meta started transferring engineers it considered "strong" into the AAI organization earlier in the month, according to the internal communications. These realignments are taking place while the company prepares a global workforce reduction: Meta plans to cut about 10% of its staff worldwide starting on May 20 and is considering further sizable reductions later in the year.
The company’s initiative reflects a broader trend in the technology sector toward automating tasks previously handled by salaried employees. Recent corporate restructuring in the United States includes significant reductions in white-collar headcount at several large firms. Amazon.com has reduced its corporate ranks by roughly 30,000 employees, a decline that has been described as nearly 10% of its white-collar workforce, while the payments firm Block said it slashed nearly half of its staff earlier this year.
Legal scholars and labor experts say that the scope of the monitoring being implemented with MCI moves data collection beyond conventional desktop logging. Yale law professor Ifeoma Ajunwa noted that while logging and screenshot tools have long been deployed to investigate misconduct or off-task conduct, recording keystrokes represents a deeper level of surveillance for white-collar employees - one more commonly associated with gig economy or delivery workers.
"On the U.S. side, federally, there is no limit on worker surveillance," Ajunwa said, adding that most state laws only require that employers provide general notice when monitoring occurs. In contrast, European legal frameworks are likely to present stronger obstacles. Valerio De Stefano, a professor at York University in Toronto who studies comparative labor law, said routines like keystroke logging would probably be unlawful under some European rules and could violate the General Data Protection Regulation in the European Union.
De Stefano pointed to examples in specific countries where courts and statutes constrain such monitoring. He said Italy expressly forbids electronic monitoring used to track employee productivity in certain contexts, and German court decisions have limited keystroke logging to exceptional circumstances, for instance when there is suspicion of a grave criminal offense. More generally, he cautioned that heightened employer monitoring tends to shift workplace power toward employers.
Meta has described MCI as a targeted data-collection effort intended to make internal AI agents more adept at emulating how people navigate software and web interfaces. The company said that the collected signals would be applied to internal model training and evaluation. Beyond Meta’s assurances of safeguards and prohibitions on using the data for employee performance measures, the memos make clear that the initiative is tied to a strategic effort to use AI agents to reconfigure how work is performed within the company.
For Meta, the stated objective is to build agents that can handle routine and complex tasks while humans focus on oversight and improvement of those agents. How regulators, employees and courts in different jurisdictions will react to the scope and methods of the monitoring remains an open question, and the company’s internal shifts coincide with large-scale workforce reductions and organizational changes aimed at embedding AI across product, engineering and operational roles.
Summary
Meta will install desktop logging software named Model Capability Initiative (MCI) on U.S.-based employee machines to record mouse movements, clicks, keystrokes and occasional screen snapshots for AI training. The company says the data will be used for model training and not for employee performance assessment. The effort forms part of a wider program, now called Agent Transformation Accelerator (ATA), to develop AI agents that perform work tasks, while Meta reorganizes roles and reduces headcount.
Key points
- Meta is deploying MCI to collect user interaction data - mouse movements, clicks, keystrokes and intermittent screenshots - from work-related applications and websites to train AI models. (Impacted sectors: Technology, Enterprise Software)
- The initiative is tied to internal workforce transformation efforts, including creation of new AI-focused teams and a push to have agents handle core tasks while humans direct and review outputs; Meta plans a global headcount reduction of about 10% beginning May 20. (Impacted sectors: Tech labor market, Corporate operations)
- Company statements assert the data will not be used for employee performance evaluation and that safeguards will protect sensitive content, though specifics on exclusions were not detailed. (Impacted sectors: Human resources, Compliance)
Risks and uncertainties
- Regulatory and legal risk in international jurisdictions - practices like keystroke logging could run afoul of European data protection laws and national statutes in countries such as Italy and Germany, potentially limiting Meta's ability to deploy the same approach globally. (Impacted sectors: Technology, Legal & compliance)
- Workplace privacy and surveillance concerns - deeper levels of monitoring may prompt employee pushback, legal challenges or reputational risk, which could affect recruitment, retention and morale in white-collar roles. (Impacted sectors: Labor markets, Talent management)
- Operational integration risk - relying on employee-generated interaction data to train agents assumes that captured signals are sufficient to create reliable autonomous systems; any gaps or misalignments could slow the intended efficiency gains. (Impacted sectors: Product development, IT operations)
Disclosure