Leaked audio shows Meta monitoring employees to train AI

The disclosure comes amid layoffs and internal unrest at Meta, where staff reportedly raised objections to mandatory monitoring software used to collect real-time work data for artificial intelligence development.

Leaked audio from an internal Meta all-hands meeting has revealed that CEO Mark Zuckerberg acknowledged and defended the company’s use of employee work activity to train its artificial intelligence systems. The meeting reportedly took place shortly before Meta began laying off thousands of employees, adding to growing internal tension within the company.

During the discussion, Zuckerberg explained that Meta’s internal system, known as the Model Capability Initiative, collects digital activity from employees’ work devices. This includes actions such as keyboard inputs, mouse movements and screen interactions across approved workplace applications like messaging tools, coding platforms and internal AI systems.

He suggested that employees were chosen as data sources because their work output is considered more advanced than that of external contractors commonly used in the industry for AI training data generation. According to him, observing internal engineers allows AI systems to learn higher-quality coding and problem-solving patterns, which could improve model performance more effectively.

Limited transparency and internal pushback over monitoring system

The Meta CEO also indicated that the company avoided publicly detailing the initiative due to competitive concerns, stating that revealing full operational strategies could benefit rivals. He emphasised that the collected data is anonymised and used solely for training AI models, not for evaluating individual employee performance.

However, internal reactions to the system were reportedly mixed. Employees expressed concern over the lack of choice in opting out of the monitoring programme. Reports suggest that a senior Meta executive confirmed there was no opt-out option for employees using company-issued devices, triggering strong reactions within internal communication channels.

The surveillance system was reportedly introduced alongside broader organisational changes, including restructuring and cost-cutting measures. These developments have contributed to uncertainty among staff, especially as Meta moved forward with large-scale layoffs affecting a significant portion of its workforce.

Rising internal discontent amid layoffs and restructuring

The rollout of the monitoring system coincided with widespread layoffs and organisational adjustments across Meta. Employees were already facing job insecurity, compensation revisions and operational restructuring when the AI training initiative was introduced, further increasing workplace anxiety.

Reports indicate that internal dissent began to surface in the form of informal protests, petitions and coordinated employee responses across multiple offices. Workers in different regions reportedly described the system as intrusive, while others raised concerns about surveillance practices in the workplace.

Zuckerberg acknowledged that communication around the initiative could have been handled better but maintained that secrecy was necessary in a highly competitive AI landscape. He also suggested that similar approaches may continue in the future as AI development becomes more closely tied to real-time workplace data.

The revelations highlight growing tension between rapid AI development strategies and employee privacy concerns within major technology firms, particularly as companies increasingly integrate internal data into model training processes.