Key facts
- Meta paused a program tracking employee computer activity for AI training due to privacy concerns.
- The tool, Model Capability Initiative (MCI), collected keystrokes, mouse clicks, and screen content.
Meta has paused a program that tracked employees' computer activity for AI training after a staff petition citing privacy concerns and reports of data exposure. The tool, Model Capability Initiative, collected keystrokes and screen content, raising questions about consent and trust.
The incident highlights the growing tension between companies' pursuit of advanced AI capabilities and employee privacy rights, potentially setting a precedent for how employee data can be ethically collected and used for AI development.
Meta has paused a program that tracked employees' computer activity for the purpose of training its artificial intelligence models, following significant employee backlash and privacy concerns. The tool, known as the Model Capability Initiative (MCI), reportedly collected data such as keystrokes, mouse clicks, and displayed screen content.
More than 1,600 Meta employees signed a petition opposing the initiative, arguing that the harvesting of 'computer use' data raised serious concerns about privacy, consent, and workplace trust. Reports from Wired indicated that data collected by MCI was accessible to anyone within the company, citing an internal security notice that detailed the exposure of sensitive information including full prompts, transcriptions, private conversations, and performance data.
Meta confirmed the pause, stating in a release, "We have carefully designed this program with privacy safeguards and while we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate." The company's founder and CEO, Mark Zuckerberg, has previously expressed that AI models learn effectively by observing skilled individuals, particularly engineers, to enhance their capabilities.
This move comes as Zuckerberg is significantly increasing Meta's investment in AI, with capital expenditures projected to reach up to $145 billion this year, largely directed towards AI infrastructure like data centers. Separately, Meta is reportedly developing a new smartphone app called Arena, which would function as a prediction market similar to Polymarket and Kalshi, allowing users to wager on future events. This venture into prediction markets, a space already facing legal scrutiny in the U.S., has been described by Forrester research director Mike Proulx as "not a great look" for a company already under legal pressure regarding its social media products.