‘Eye of Sauron’, ‘super uncomfortable’: Why did Meta employees revolt against its AI experiment?

Meta’s ambition to build more powerful AI hit an unexpected roadblock, and that is not from the competitors, but from its own employees.

Recently, the company introduced an internal programme called the Model Capability Initiative (MCI), specifically designed to collect data from tens of thousands of employees to help train AI models. It tracked details such as keystrokes and mouse movements, allowing AI to learn how humans operate computers, including their everyday actions. But the entire initiative faced immense resistance from employees.

What did the employees protest about?

When Meta’s chief technology officer, Andrew Bosworth, said employees would not be allowed to opt out, the top-rated response was, “This makes me super uncomfortable.”

Meta pointed out that the data would be protected by strict privacy controls. And responding to concerns like information leaks, Bosworth added, “That isn't remotely a fair characterisation! How would any of this end up in ‘the open’? This is tightly controlled data.”

Employees still remain sceptical about what the programme could actually do. They worry that screen data can easily capture information such as passwords, immigration forms and credit card details. Some even worried that the same data being collected to teach AI how humans work could eventually replace them.

One of the employees who opposed the programme called it a “big reality check”. Some of them began sharing anti-AI memes and described the screen-recording icon as the “Eye of Sauron”.

More than 1,800 employees eventually signed a petition demanding that the programme be stopped.

Meta even tried to make some changes to the programme, such as allowing employees to temporarily opt out of sensitive tasks for up to 30 minutes. But they rejected the move as “insufficient and unsatisfactory”, demanding that Meta scrap MCI “entirely, immediately, forever.”

Why was MCI shut down?

The biggest blow came in June, when an internal leak report showed that MCI had collected several sensitive data points, including private conversations, and they were accessible across the company.

One employee wrote that they were “incensed”, while another sarcastically asked Meta’s AI assistant: “Hey Metamate, define the word ‘hubris’ for me.” Meta shut down the programme within hours.

Meta said MCI was suspended, rather than cancelled, and would become optional if revived. The episode highlights a growing problem for AI companies: the data needed to teach machines how humans work may increasingly come from humans who do not want to be watched.

William Fitzgerald, founder of The Worker Agency, called the employee pushback “a historic win”, saying: “Zuckerberg wanted more data — he tried to do it this way, but it didn't work."