Meta needed employee data for AI edge, but faced a very 'human problem'

Meta needed data from thousands of its employees for edge in AI space, company introduced Model Capability Initiative; but then came a very 'human problem'

Meta found an internal way to get its hands on more data to strengthen its position in the artificial intelligence (AI) race. To bridge the data gap that didn't come from the internet, the social media giant was turning to its own employees.

Meta’s elite AI unit, known as TBD, introduced the Model Capability Initiative (MCI), a programme designed to collect data on how thousands of employees use their computers.​The idea was simple: keystrokes, mouse movements and screen activity could help AI models learn how humans actually work with computers. But Meta soon ran into a very human problem: opposition from its own employees. What followed was a rebellion involving memes, flyers and a petition signed by more than 1,800 workers.​

Why Meta wanted employee data

​MCI was ‘invented’ by Meta because of the fundamental drawbacks that their AI systems had. Although the AI software is skilled in certain domains such as programming and research, it has not been able to perform basic computer tasks.​According to people familiar with the project, Meta believed that training AI on employee data would help the systems understand how people actually do things. MCI was conceived in TBD, an AI research lab of about 100 people, which is seen as one of Meta's most advanced AI labs.

​

Employees immediately pushed back

​When Meta announced MCI internally in April, the response fell far short of what the company expected.​Employees raised concerns that the programme would capture sensitive information from their computers, including passwords, immigration documents and credit card details. Some also worried that the data could eventually be used to automate jobs currently performed by Meta employees.​The backlash quickly spread through internal channels. Employees shared anti-MCI memes, while the recording icon was mockingly referred to as the “Eye of Sauron”.​An internal petition opposing the programme eventually collected more than 1,800 signatures. Employees also distributed flyers at Meta offices and put up messages protesting the initiative.​At first, Meta gave workers the option to opt out for 30 minutes in June so they could do personal business. Nonetheless, activists found it insufficient and demanded that Meta cancel MCI.

​

Then came the data leak

​However, MCI took its biggest blow toward the end of June. It emerged that data collected using the software was visible throughout Meta and even included private conversations.

The leak became an instant viral sensation within Meta, seemingly confirming employees' fears about tight control of the data.​Meta subsequently switched off MCI within hours. The company said it had suspended the programme and would make it opt-in if it brought it back. Two months later, it remained suspended.​Uncertainty also remained over whether the collected data had actually been used to train Meta’s latest AI models. A manager reportedly said in July that it had not been used, while Meta CTO Andrew Bosworth told employees that the data was more useful than expected.​