AI Companies Really, Really Want to Buy Your Work Files and Emails

The personal data from your work emails and documents can now be sold to AI companies, potentially becoming a significant revenue stream for your employer.

AI companies are aggressively acquiring vast amounts of private work data, including emails, call transcripts, and internal documents, to train their models, a market that industry insiders estimate is already worth tens of billions annually and could exceed a trillion dollars in the next few years. Companies like Handshake, Mercor, and Micro1 are actively purchasing this data, often from businesses undergoing wind-downs or directly from individuals, promising anonymization. Mercor, founded in 2023, has already produced billionaire co-founders, highlighting the rapid growth in this sector. The practice has drawn scrutiny, notably in the case of a bankrupt Spirit Airlines attempting to sell its digital assets, which include millions of internal messages and emails, facing objections from the flight attendants' union over potential privacy violations and the use of worker data to develop AI that could automate their jobs. Major AI labs like Anthropic acknowledge the data market but are cautious about directly purchasing live company workflows, while OpenAI has not yet responded.

The burgeoning market for AI training data signifies a critical juncture where the value of human-generated content is being rapidly commodified, with companies actively sourcing private corporate workflows previously considered internal. This trend highlights a significant shift in data economics, moving beyond publicly available information to proprietary corporate assets, driven by the insatiable demand for more sophisticated AI models. The rapid emergence of well-funded startups and the quick accumulation of wealth by their founders underscore the immense, largely untapped financial potential perceived in this data acquisition sector.

This data gold rush poses substantial ethical and privacy challenges, particularly concerning employee consent and the potential for de-anonymized data leakage, as evidenced by union objections to corporate data sales. As AI models are trained on this data with the explicit aim of automating human tasks, it creates a paradoxical situation where workers' own professional output contributes to the technology that may displace them. The lack of established industry norms and regulatory oversight suggests a "Wild West" environment, where the speed of AI development outpaces the creation of responsible data governance frameworks.