More Perfect Union, a workers’ rights advocacy group, posted an audio clip on X on May 20 allegedly from Meta’s all-hands meeting held on April 30. In the recording, a voice believed to be Mark Zuckerberg responded over six minutes to employees’ questions regarding ‘device monitoring’. The speaker stated that Meta collects employees’ keystrokes, mouse clicks, and screenshots, feeding them into AI models ‘so they can learn how smart people use computers’. He also explained why Meta prioritizes monitoring its own staff rather than contractors: ‘The average intelligence level of our employees is significantly higher than that of ordinary individuals obtainable through outsourcing firms.’ Meta has yet to publicly confirm or deny the authenticity of this recording.
The timing of this leak has sparked even greater public outcry: around the same time, Meta announced layoffs affecting roughly 8,000 employees. Critics have drawn a direct link between these two events, arguing that the company is using employees’ work behavior data to train AI systems designed to replace them. In January this year, Wired reported that Handshake AI, a data provider for OpenAI, required outsourced contractors to upload actual work documents—including contracts, financial models, and code repositories—for AI training; intellectual property lawyers have since warned of potential legal risks. Together, these incidents highlight a growing trend in the tech industry: the systematic conversion of high-IQ white-collar workers’ behaviors into assets for AI training.