In 'leaked' audio from Meta townhall, CEO Mark Zuckerberg tells employees: Meta intentionally kept staff in the dark because…

Days before Meta started laying off 8,000 people, Mark Zuckerberg gathered his employees for an all-hands meeting and told them something the company had avoided saying publicly: their computers were being used to train AI, they were chosen specifically because they're smarter than outside contractors, and no—Meta didn't explain any of this upfront because it didn't want rivals finding out.

In one meeting, Zuckerberg managed to confirm the surveillance, justify it, and defend the secrecy around it, more or less in that order.The audio, obtained by More Perfect Union from the April 30 all-hands, captures Zuckerberg explaining Meta's Model Capability Initiative—the mandatory software that logs employee keystrokes, mouse movements, and screen activity—in his own words. The explanation he gave internally is blunter than anything Meta said publicly: the company chose its own employees as AI training subjects because they're smarter than the contractors the rest of the industry uses.

And they didn't spell that out because they didn't want competitors to catch on."It is not strategically in your interest for us to communicate everything in all the detail that we normally would on this," Zuckerberg said.

Why Meta is watching its own employees—and not outside contractors

For years, AI companies have relied on contract workers to generate training data—clicking through interfaces, labeling images, annotating text. Meta is doing something different. It's harvesting that data from its own engineers, live, as they work.

Zuckerberg's rationale is straightforward, if a little pointed. "The average intelligence of the people who are at this company is significantly higher than the average set of people you can get to do tasks if you're working through these contractors," he said in the leaked audio. If you want AI to learn how to code like a senior engineer, have it watch senior engineers code. "Having people internally build tools or solve tasks that help teach the model how to code, we think is going to dramatically increase our model's coding ability faster than what others in the industry have the capability to do.

"The MCI tool—first reported by Reuters in April—runs across a pre-approved list of work apps: Gmail, GChat, VSCode, Meta's internal AI assistant Metamate. It captures mouse clicks, keystrokes, and periodic screen snapshots. Zuckerberg said the content is stripped of identifying information and that no human is watching. "None of the data has been used for looking at what people are doing or surveillance or performance tracking," he said.

"It's purely just like we are using this to feed a very large amount of content into the AI model."Employees were not given the option to disagree. When the announcement landed internally, the top comment was: "This makes me super uncomfortable. How do we opt out?" Meta CTO Andrew Bosworth's reply, reported by Business Insider, was one line: "There is no option to opt out of this on your work provided laptop." It received over 100 angry and shocked emojis in response.

Inside a company on the edge: protest flyers, petitions, and a union drive

The MCI rollout didn't happen in a vacuum. It landed the same week Meta confirmed it would cut 10% of its global workforce—the layoffs began rolling out this morning, May 20. Employees were already rattled by mandatory reassignments, two consecutive years of stock-based pay cuts, and a median total compensation that dropped from $417,400 in 2024 to $388,200 last year, according to Wired.Into that atmosphere, a mandatory surveillance program arrived with minimal explanation.

"Layoffs, budget cuts, years of efficiency and intensity—all of it contributed to a growing sense of dread," an engineer wrote in an internal post seen by nearly 20,000 colleagues, per Wired. "MCI is a microcosm for the AI movement."Protest flyers went up in cafeterias, bathrooms, and on vending machines across US offices, pointing employees to a petition demanding the program be shut down—Reuters first reported it.

In the UK, workers began organizing with United Tech and Allied Workers, calling it "draconian surveillance." The New York Times reported employees created countdown websites to the layoffs, one of them titled, bleakly: "Big Beautiful Layoff.

"Zuckerberg acknowledged the rollout was poorly handled. "There are all these things that could have been done better. Yes, acknowledged," he said. But he stood by the secrecy.

AI is too competitive a space to explain strategy in full, he said, knowing it will leak.

'This probably isn't the last thing like this'

Before wrapping up the topic, Zuckerberg offered what may be the most clarifying line of the whole meeting: "This probably isn't the last thing like this."Meta is forecast to spend between $125 and $145 billion on AI infrastructure this year alone. The layoffs landing today are, in Gale's own framing, meant to "offset" that investment. The employees who remain will be fewer, more monitored, and working inside a company that has now said—on tape—that their daily work habits are a competitive asset it intends to keep mining.