Meta has been exposed for continuously screen-recording employees to train AI agents, while simultaneously moving forward with 8,000 layoffs—your very own AI, which you helped teach, may now be replacing you.Article author and source: AI New Era
In the morning, a Meta employee turned on their work computer as usual.
He noticed that the system had an extra component installed.
No pop-up notifications, no permission requests. A program called MCI has been logging every mouse movement, every click, every keystroke, and occasionally taking screenshots.
Only a casual note in the internal memo: “This tool will record your mouse movements, clicks, and keystrokes within work applications.”
He froze for three seconds.
It wasn’t because of the monitoring itself—doesn’t every major tech company in Silicon Valley monitor its employees? It was because he suddenly remembered another line from the all-hands email two weeks ago: “Starting May 20, the company will eliminate approximately 8,000 positions.”
The internal memo obtained by Reuters clearly states that MCI collected data to train AI agents capable of "autonomously performing tasks."
Tech media Business Insider reported that the program had no opt-out option, and Meta employees were collectively forced to be "refined" by AI!
Zuckerberg's "Digital Slaughterhouse"
Even more strikingly, Business Insider exposed the full memo:
MCI, short for Model Capability Initiative.
One sentence in the memo stood out: “This is precisely the opportunity for all Meta employees to help improve our models through their everyday actions.”
In plain terms, your work itself has become training data.
Essentially, it is an extremely sophisticated "soul-transfer device."
It no longer settles for the basic "attendance monitoring" of the past, but delves into the micro-level of human cognition.
When you're troubleshooting a complex ad campaign bug, MCI records not just the outcome, but how you clicked, how you stepped back, and how you compared information across three windows.
Why does Meta need your "clickstream"?
In the logic of AI training, synthetic data is facing model degradation due to "inbreeding."
Scientists at Meta SuperIntelligence Labs understand that the most scarce causal chains are found only in human workflows.
Each keyboard shortcut represents a logical decision; each mouse movement reflects an intuitive judgment.
Meta, through MCI, is converting these "tacit knowledge" elements—previously existing only in the human brain—into structured data that AI can ingest, via live recording.
This is an extremely efficient form of "behavioral cloning."
"AI Builders": A New Species Is Born
Meta is eliminating traditional functional boundaries.
Engineers, product managers, and designers—these titles are being replaced by a single generic term: "AI builder."
Last month, Meta established a new Engineering team called "Applied AI," dedicated to enhancing the coding capabilities of AI models, with the goal of enabling agents to handle most tasks related to product and infrastructure development, testing, and deployment.
At the beginning of this month, the company began transferring senior software engineers to this team.
Note this screening criterion—only “strong” engineers qualify. What about the rest?
On Monday, Meta's Chief Technology Officer, Andrew Bosworth, sent an internal memo formally renaming the original "AI for Work" initiative to the "Agent Transformation Accelerator" (ATA).
He wrote: “The vision we are building is for agents to primarily do the work, while our role is to direct, review, and help them improve.”
Command, review, improve—pay attention to these three words. Previously, humans did the work and AI was a tool. Now it’s reversed: AI does the work, humans review, but how long will this reviewing role last?
On May 20, Meta will initiate a global workforce reduction of 10%, approximately 8,000 employees.
Previously, Zuckerberg predicted:
We will reach a stage where all the code in our applications, as well as the AI it generates, will be written by AI engineers rather than human engineers.
At this moment, connecting what Meta has done is truly chilling.
A Meta spokesperson emphasized that MCI data "will not be used for performance evaluations," is solely intended for model training, and that protective measures have been implemented to ensure the security of "sensitive content."
This is true in a sense—
Because once you’ve been drained of value as a “teaching sample,” your performance becomes meaningless.
The company is no longer evaluating your performance; it is replicating your existence.
In the digital slaughterhouse, employees are not crops to be harvested, but raw materials to be refined into biofuel.
It's not just Meta going crazy.
In recent months, Amazon laid off 30,000 corporate employees, about 10% of its white-collar workforce. Fintech company Block was even harsher—cutting nearly half of its staff.
By the second quarter, Amazon will also lay off tens of thousands of employees, and Silicon Valley cloud giant Salesforce will cut tens of thousands of jobs as well. Even more alarming, these layoffs are just a warm-up—many more companies may appear on the 2026 layoff watchlist.
Official Layoff, which tracks layoffs in real time, made a startling claim: "In one year, layoffs will sweep across all jobs worldwide!"
The world is watching: Is your country protecting you?
Emory University law professor Ifeoma Ajunwa bluntly stated: “There are no federal restrictions on employee monitoring in the United States.” State laws require employers to provide only “general notice.”
Meta’s approach—notification via internal memo—technically meets this extremely low threshold.
Across the Atlantic, it’s a different story.
York University law professor Valerio De Stefano noted that European law is likely to prohibit such monitoring.
In Italy, using electronic monitoring to track employee productivity is explicitly illegal. In Germany, keystroke logging is permitted only when there is suspicion of serious criminal activity.
He said something worth every worker remembering: the very existence of employer surveillance shifts the balance of power in the workplace.
You don't actually need to be replaced by AI.
You just need to know that a program is logging every action you take, and your boss can use this data at any time to train a digital clone that doesn’t require a salary, doesn’t need rest, and never complains.
This very awareness has already turned you into a more obedient component.
Your transaction history is your epitaph.
The memo from Meta also stated: "The goal is for agents to automatically discover where we have intervened, so we can do better next time."
Read it again.
It’s not that AI learns your job and stops. Rather, AI remembers where you corrected it and learns to no longer need your corrections.
Every time you teach AI to do better, you get closer to being replaced.
The cruelest thing isn't unemployment—it's how you lose your job.
You might say: I don’t work at Meta, so why should I care?
Wrong. Meta was just the first to be exposed.
Microsoft's AI capital expenditure this year: $80 billion. Google: $75 billion. Amazon: $100 billion.
Not all of this money went toward training large models. A significant portion was spent on "learning how humans work."
Anthropic has released Claude's Computer Use feature, which enables control of a computer.
OpenAI's Operator can automatically complete browser tasks.
Google's Project Mariner enables AI agents to work directly within your Chrome browser.
Exactly aligned: Teach AI human workflows, then let it take over.
For the first time in human history, "working hard" and "keeping your job" point in completely opposite directions.
Reference materials:
https://www.reuters.com/sustainability/boards-policy-regulation/meta-start-capturing-employee-mouse-movements-keystrokes-ai-training-data-2026-04-21/
https://www.businessinsider.com/meta-cto-andrew-bosworth-reality-labs-all-hands-meeting-2026-1
https://www.aol.com/articles/meta-track-workers-clicks-keystrokes-224800463.html
https://x.com/Reuters/status/2046636232554234133?s=20