Employee Data for Agent Training
- Podcasters and YouTube reporting say Meta is logging employee clicks, keystrokes, and screenshots to train office-task AI agents. - That collection aims to improve agent performance but raises clear privacy, labour, and legal exposure for employers. - The coverage recommends auditing monitoring practices now and warns such internal data strategies could trigger backlash and regulatory scrutiny (youtube.com) (youtube.com).
Meta is rolling out software that records U.S. employees’ clicks, keystrokes and some screenshots to train artificial intelligence agents that can use workplace software. (reuters.com) Reuters reported on April 21 that Meta told staff in internal memos it would install the tool on work computers as part of a “Model Capability Initiative,” or MCI. The system is designed to capture mouse movements, clicks and keyboard input from U.S.-based employees and contractors. (reuters.com) CNBC reported on April 22 that Meta’s monitored list includes hundreds of sites and apps, including Google, LinkedIn and Wikipedia, based on internal documents it reviewed. Reuters and CNBC both said the software is limited to work-related apps and websites, not personal browsing on unmanaged devices. (cnbc.com) The underlying problem is simple: software “agents” need examples of how people move through forms, menus and search boxes to finish office tasks. Meta told employees the data would help models learn how humans actually navigate digital tools, according to Reuters and TechCrunch. (reuters.com) (techcrunch.com) That turns ordinary office work into training material. Ars Technica, citing the Reuters report and internal memos, said the system also takes periodic screenshots to give the models more context about what was on screen when a worker clicked or typed. (arstechnica.com) Meta said participation would not be mandatory for everyone. Reuters reported that employees handling especially sensitive work, including some legal and human-resources functions, could seek exemptions from the program. (reuters.com) The legal exposure is not hypothetical. New York’s electronic monitoring law, in effect since May 7, 2022, requires private employers with a place of business in the state to give written notice when they monitor employee phone, email or internet use. (newyork.public.law) Employment lawyers have warned in 2026 that keystroke logging, screenshot capture and device monitoring can create privacy, consent and record-retention risks even when employers own the hardware. Nelson Mullins said companies using electronic monitoring should document purpose limits, notice practices and data-handling rules before expanding surveillance. (nelsonmullins.com) Europe is tightening the rules too, though on a different slice of workplace AI. The European Commission’s guidance on the Artificial Intelligence Act says emotion-recognition systems in workplaces are prohibited except for narrow medical or safety uses, a sign that regulators are drawing lines around worker monitoring. (digital-strategy.ec.europa.eu) (iapp.org) Meta’s program is narrower than emotion recognition, but it points the same direction: companies want proprietary human-behavior data to build more capable workplace automation. The next fight is likely to center on notice, consent, exemptions and whether employees’ daily work can be repurposed as training data without a broader bargain. (reuters.com) (cnbc.com)
