OpenAI, Anthropic, Meta and Google race to buy workplace records as AI training fuel; consent dispute grows

As OpenAI, Anthropic and Meta develop AI agents to automate white-collar work, internal corporate data is being bought and sold as AI training fuel, with employee consent and data-protection disputes growing, Digital Today reports. Business Insider reported (Oct. 13) that AI data brokers are buying workplace records not only from shuttered startups but also from companies still in operation, in some cases presenting offers worth hundreds of thousands of dollars that required a decision within 48 hours. Key deal targets include Slack conversations, emails, customer-management records, presentation materials and source code — records that capture employees' decision-making processes and can train AI agents to perform complex office tasks. Brokers say they strip identifiers such as social security numbers, but in some cases employees do not know their work records have been sold to outside parties.

Meta sought to use employees' keystrokes and mouse movements for AI training and halted the project after it emerged the anonymised data included private conversations and performance review records. Google's plan to buy past data from Spirit Airlines is also being delayed over concerns about the use of employees' personal information. Emory University law professor Ifeoma Ajunwa noted it is "ethically unclear whether consent to collect employee data permits external sales as well." Market researcher Gartner forecasts global spending on data for AI will rise to $6.48 billion in 2027 from $826 million in 2025. All allegations are based on reporting; no legal findings have been made.