The scariest corporate story of 2026 isn't a flashy slide deck. It's your ordinary Tuesday at your desk.
Press including the BBC and CNBC reported on Meta's Model Capability Initiative (MCI):
- Meta built an internal tool to capture keystrokes, mouse clicks and navigation on work machines.
- Across hundreds of sites — Google, LinkedIn, GitHub, Slack, Wikipedia and more.
- The stated goal: train AI agents that use a computer the way a person does — "real examples of how people actually use them."
- Employees called it "dystopian" in internal messages — especially set against the fear of job cuts.
- After the backlash: a 30-minute pause option and some exemptions (Reuters, Jun 2026).
Meta isn't unusual in wanting behavioural data — they're just the headline example you can name down the pub. And the principle generalises: every email you draft, every ticket you close, every spreadsheet you nudge is a training signal the moment your employer logs it for AI.
So you're not being paranoid if you ask:
- Is my company recording my screen or my apps to train a model?
- Am I quietly speeding up a tool that's designed to replace part of my own role?
- Who actually owns the model that gets built out of my work?
The partial defence is simple to say and harder to do: get yourself into the group that directs, audits and deploys AI — not just the group whose output feeds it blind. That, in one sentence, is why I built this course.
Continue — the physical robots.