It didn’t end well for the surveillance machine.
Meta’s Model Capability Initiative, the controversial system that logged employee keystrokes and mouse movements to train its AI, has been paused. The tool launched in April. It faced immediate resistance. Then things went sideways.
Sensitive internal data got exposed.
The leak
Private conversations. Prompts. Performance reviews. Transcriptions. According to Wired, all of it became accessible to anyone inside Meta. An internal security notice flagged the breach, corroborated by three employees who saw what happened.
This isn’t a small thing. Over 1,600 employees signed a petition demanding an end to the collection and repurposing of their computer data. Engineers, researchers, designers — the people actually building the products.
Their argument was straightforward. You can’t build “responsible AI” while ignoring basic human boundaries.
“Any approach to AI that relied on intrusive, coercive non-consensual data collection contradicted that principle,” the petition stated.
The software tracked everything. Gmail. Google Chat. Even Metamate, the internal AI assistant. It took screenshots, too. A Meta spokesperson claims there are no indications yet that the data was improperly accessed, despite the exposure. Still. They are investigating. Indefinitely halting the program for now.
Did it have to happen this way?
Big money, bigger ambition
Context matters here. Meta isn’t doing this because it’s paranoid. It’s doing this because they are pouring $135 billion into AI infrastructure this year alone. Amazon is up to $200 billion. Microsoft sits at $190. Alphabet is at $185.
This is an arms race.
In leaked audio from April 30, Mark Zuckerberg argued that his employees were simply too good to waste on low-level tasks. He believed the company’s average intelligence was significantly higher than external crowds for task completion. The logic was that AI models learn faster by watching smart people work.
Efficient. Maybe. Also, deeply unsettling.
The human cost
Rory Mir from the Electronic Frontier Foundation calls it an abuse of power. Seeking new training data isn’t an excuse for disproportionate monitoring.
“It highlights the necessity of legislation,” Mir said, demanding consent and due process.
Right now, those laws aren’t in place. Meta is pausing. But the pressure to grow won’t stop. The capital is spent. The ambition remains.
Whether this tool comes back or something worse takes its place is unclear. For now, the logs are stopped.
But the question isn’t whether companies will try to watch us.
