Meta has paused its “Model Capability Initiative” — a program that tracked employee keystrokes and mouse clicks to harvest behavioral data for AI training — after discovering the collected data had been left potentially accessible to anyone inside the company. Nearly 2,000 Meta workers signed a petition demanding the program be cancelled. The halt comes as Meta spends $145 billion on AI this year while morale inside the company hits what employees describe as its lowest point ever.
🔍 THE BOTTOM LINE
Meta’s retreat from MCI is damage control following a security lapse that exposed worker surveillance data internally. The program — started just two months ago — tracked every mouse click and keystroke from employees to build training datasets for AI models. When that data was found to be accessible company-wide, it crystallized months of employee anger. With Meta already bleeding talent from its “atrocious” AI reorg and spending $145bn to catch OpenAI and Anthropic, the pause reveals a company where the gap between AI ambition and workforce trust has become unbridgeable.
What Went Wrong
The MCI program launched just two months ago as part of Meta’s push to gather proprietary behavioral data — the kind of granular, human-computer interaction data that could give its models an edge over competitors. Employees were tracked continuously: mouse clicks, keystrokes, application usage patterns, all logged and stored.
The problem was twofold. First, the tracking was mandatory and forced on workers without meaningful consent. According to BBC News, employees immediately pushed back, with nearly 2,000 signing a petition demanding the program be cancelled. Meta’s initial response — allowing workers to pause tracking for 30 minutes at a time — was described by one employee as “just an attempt at damage control.”
Second, the data itself was mishandled. Meta halted the program on Monday after realizing some of the collected data had been left potentially accessible to anyone inside the company. A Meta spokesman confirmed to the BBC that MCI was “on pause for now” while adding: “We have no indication at this time that any data was improperly accessed by Meta employees.”
The Broader Morale Crisis
The tracking program didn’t emerge in isolation. It follows months of internal turmoil at Meta, where CTO Andrew Bosworth reportedly called the company’s AI reorganization “atrocious” — a process that displaced 6,500 engineers and gutted morale. Employees have openly insulted management using explicit language in internal meetings, according to a Wired report.
The company is spending up to $145 billion on AI this year alone, even as it conducts extensive layoffs and reorganizes teams around AI initiatives. One current employee told the BBC: “I’ve never seen morale here so bad.” Another described the direction as “depressing” and “exhausting.”
The contradiction is stark: Meta is spending more on AI than any company in history while simultaneously alienating the workforce it needs to build the products. Workers who might have voluntarily contributed to AI training data collection are instead being surveilled without consent — and then told the data wasn’t even properly secured.
NZ Angle
For New Zealand, the Meta MCI case is a warning shot. NZ’s Privacy Act 2020 requires explicit consent for workplace monitoring, and the Employment Relations Authority has previously ruled against covert surveillance. But as AI-hungry companies look for behavioral training data, the pressure to normalize always-on workplace tracking will increase.
Kiwi employers adopting AI tools should note: Meta’s approach — forced, opaque, insecure — is the exact opposite of what a consent-based framework looks like. The lesson isn’t that behavioral data collection is impossible; it’s that doing it without trust, transparency, and security guarantees a backlash that costs more than the data is worth.
The Bigger Picture
Meta’s pause is the first major instance of a frontier AI company being forced to retreat from internal surveillance by its own workforce. The industry has been quietly harvesting employee data for model training for years — Meta just did it clumsily enough to get caught.
The broader question is whether AI companies can sustain the data-harvesting model their technology depends on. Synthetic data and consent-based collection exist as alternatives, but neither delivers the volume or fidelity of forced workplace surveillance. If employees at every major AI lab push back the way Meta’s did, the data pipeline that fuels model improvement narrows.
❓ FAQ
What was the Model Capability Initiative (MCI)? A Meta internal program that tracked employee computer usage — mouse clicks, keystrokes, application usage — to generate behavioral data for training AI models. It started roughly two months ago and was mandatory for affected workers.
Why did Meta pause it? The company discovered that collected data had been left potentially accessible to anyone inside Meta. The pause follows weeks of employee backlash, including a petition signed by nearly 2,000 workers.
How much is Meta spending on AI? Up to $145 billion this year, including infrastructure buildout, model training, and AI-related reorganization costs.
Could this happen at a NZ company? Workplace surveillance without informed consent violates NZ’s Privacy Act. But as AI companies seek behavioral training data, pressure to normalize tracking will grow. The Meta case shows what happens when consent and security are both skipped.
🔍 THE BOTTOM LINE
Meta’s pause of its worker-tracking program is less a moral awakening than a forced retreat — the data was insecure, the workforce was in open revolt, and the company’s AI reorg had already destroyed the trust needed to implement such a program voluntarily. The $145bn AI spend continues, but the human cost is now visible: a workforce that describes its own morale as the worst it’s ever been, a tracking program that couldn’t even keep its surveillance data secure, and a company that has proven it can’t pursue AI dominance and worker trust at the same time.
