Memorandum
- From
- Derek Weston via Fast Company
- Date
- Filed
- Business·4 min to read
- Re
Meta Shelves AI Data Collection Program After Employee Backlash
ReMeta Shelves AI Data Collection Program After Employee Backlash
Meta ended its Model Capability Initiative after employees revolted over mandatory data collection and a privacy breach exposed company-wide access to private conversations.
Meta has suspended an internal program designed to collect employee data for training artificial intelligence models after a swift and vocal backlash from its workforce, according to a new report. The program, called the Model Capability Initiative (MCI), aimed to capture keystrokes and mouse movements from employees to improve AI systems. But when workers learned the data collection was mandatory and offered no way to opt out, they organized quickly, circulating a petition that ultimately gathered more than 1,800 signatures and distributing physical flyers to rally support.
The pushback began almost immediately after the program was announced. Employees raised concerns that the data collection could inadvertently capture sensitive information such as passwords, credit card details, and immigration forms. A cybersecurity engineer warned that security appeared to be «an afterthought,» prompting Meta CTO Andrew Bosworth to push back sharply. «That isn’t remotely a fair characterization! How would any of this end up in ‘the open’? This is tightly controlled data,» Bosworth said, according to the report. Despite his assurances, anxiety spread among staff, with some fearing the program could be used to replace human workers with AI.
The idea for MCI originated within Meta’s AI division, which operates largely independently from the rest of the company. One of the researchers leading the effort was Rylan Schaeffer, who reported to Meta’s AI chief. The program’s mandatory nature and the lack of an opt-out option fueled the outrage. In June, Meta attempted to quell concerns by allowing employees to opt out for up to 30 minutes when handling sensitive matters, but workers continued to demand that the company abandon MCI entirely.
Those privacy concerns proved justified. Toward the end of June, a Meta engineer discovered that data collected by the program was accessible to the entire company, including private conversations. The revelation triggered immediate action. Meta suspended the program almost instantly and said employees would be able to opt out if it were ever reintroduced. One employee reportedly summed up the sentiment: «I swear we’re probably the richest yet most incompetent company to exist in tech right now.»
Conflicting accounts have emerged about whether the collected data was actually used for AI training. Employees told Business Insider that one executive claimed the data was not used, while Bosworth reportedly said it proved more useful than anticipated. In leaked audio obtained by More Perfect Union, Bosworth maintained that employees who openly opposed MCI «made the entire company worse.» Meta declined to comment when reached by Fast Company.
The episode highlights growing tensions between tech companies racing to build AI and white-collar workers who are increasingly willing to voice objections to workplace monitoring. Unlike typical monitoring of hourly or warehouse workers, this program targeted salaried employees who are not usually subject to such surveillance. The backlash also comes amid a broader wave of AI-driven layoffs across the tech industry, with many workers fearing they could be replaced by the very systems they are asked to help train.
Meta’s experience suggests that even legally permissible data collection on company devices can carry significant reputational and internal risks. The rapid retreat underscores how quickly employee dissent—amplified by petitions, leaks, and public scrutiny—can force a major corporation to reverse course. As the AI boom continues to reshape the tech labor landscape, the MCI episode may serve as a cautionary tale for companies seeking to harvest worker data in the name of innovation.
