Meta Pauses MCI After Internal Revolt and Data Leak

Key Takeaways

1Meta has suspended MCI, an internal program capturing keystrokes and screens to train AI, after an organized protest and the discovery that data was visible internally.

2A petition gathered over 1,800 signatures, and activists claim to have ensured that any potential return would be voluntary.

3Management defends the usefulness of the data while stating that it has not been used for its latest models, and maintains the program on pause.

Why it matters : The case illustrates internal resistance and the risks associated with collecting work data for AI training.

Meta has suspended MCI, a program aimed at capturing keystrokes and screens of employees to train its systems, following an internal leak and organized protests. Executives assert that the data was useful and under control, while employees cite sensitive risks and claim victory. Two months later, the program remains halted and would only return on a voluntary basis, according to the company.

Program Disabled After Internal Data Visibility and Public Criticism

After an internal leak during the summer, Meta suspended the MCI program and stated that it was not canceled but deactivated, with the prospect of optional operation. Two months later, the company confirmed that the program remained on hold. Meta declined to comment on how, or if, the data had been used, while an executive claimed in July that this data had not been used to train the latest models. During a meeting that same month, Andrew Bosworth stated that the data had proven to be more useful than expected. In a leaked recording, he repeated three times that opponents of MCI had made the company worse and argued that the petition had complicated internal developments.

Petition, Memes, and Labor Rights: A Structured Mobilization at Meta

The protest took the form of a petition signed by over 1,800 employees, anti-AI memes circulated internally, and actions on campuses, including the distribution of flyers mentioning protections under the National Labor Relations Act. The link to the petition was even posted on a building at the Burlingame site. In June, Meta conceded a 30-minute window allowing employees to temporarily opt out of tracking, a measure deemed "insufficient and unsatisfactory" by activists, who demanded the complete and permanent abandonment of the program. Opponents believe their actions led Meta to consider a potential return of the program only on a voluntary basis and describe the outcome as the result of "courageous acts of resistance."

Internal Discovery in Late June: MCI Data Visible Company-wide

At the end of June, an engineer discovered that information collected by MCI, including complete private conversations, was accessible to the entire company. Employees expressed outrage and recalled previous assurances regarding the security of MCI data. Employee comments mocked the situation, referencing "hubris" directed at the internal AI assistant and criticizing the company's competence.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Employee Concerns: Sensitive Data and AI Replacing Evaluations

Some organizers reported concerns about the potential collection of confidential information, such as passwords, immigration forms, or credit card numbers, from work terminals. Others expressed fears that these captures could be used to replace employee evaluations with AI tools, a danger described as a "rude awakening." The icon displayed during screen recording was ironically dubbed "Sauron's Eye" by employees.

Genesis and Objectives of MCI: A Program Born at TBD and Announced in April

MCI originated within TBD, a unit of about 100 researchers known for its high autonomy within the Menlo Park campus. One morning in April, the initiative was presented to the entire workforce: the announcement emphasized that the models were still struggling with basic computer usage, such as menu selection and shortcuts. The reaction was immediate, with numerous angry emojis, after CTO Andrew Bosworth dismissed any possibility of opting out. The most liked comment read, "This makes me very uncomfortable." Bosworth defended the program's confidentiality, responding to a cybersecurity engineer who questioned data security and criticized a late security approach, assuring that the data was "strictly controlled." The MCI aimed to capture keystrokes and mouse movements to help systems replicate human usage and was intended to be mandatory.

Decision-Making Chain and Personnel Movements Around the Project

According to three interviewed individuals, Rylan Schaeffer was among the leaders of MCI and reported to Alexandr Wang and Jack Rae, described as at the top of Meta's AI efforts. According to his LinkedIn profile, Schaeffer left the company in July for a new project. The previous spring, members of TBD had sought a competitive advantage by relying on data from tens of thousands of employees, a direction that triggered internal opposition manifested through memes, flyers, and the petition.

Beyond Meta: Increased Caution Among Companies Regarding Work Data

External precedents illustrate resistance to the commodification of work data, such as the action taken by former flight crew members of Spirit Airlines who temporarily blocked a $10 million database sale to Google. William Fitzgerald from The Worker Agency anticipates that the reaction at Meta will prompt other companies to hesitate before imposing data sharing for training, calling it a "historic victory" and a failed attempt at data collection desired by Mark Zuckerberg. Meta stated that any potential return of MCI would be voluntary and confirmed, two months after the suspension, that the program remained on hold, declining to comment further.