AI’s New Training Data: Your Old Work Slacks And Emails

The operational data of defunct companies is emerging as a valuable, unexpected asset for training the next generation of Artificial Intelligence. Shanna Johnson, former CEO of transcription company Cielo24, discovered that the company's 13-year digital footprint, including Slack messages, JIRA tickets, and internal emails, could be sold as training data for AI. This sale generated hundreds of thousands of dollars, providing a clean financial exit for the company's closure. This trend represents a new frontier in the AI industry. After exhausting publicly available data like internet forums and books, AI labs now require more specific, "handcrafted" data from real-world work environments to train "agentic AI" – models capable of performing tasks. The internal communications and operational records of companies like Cielo24 serve as a rich source of this practical, work-related data. Companies like Simple Closure are capitalizing on this demand. Their CEO, Dori Yona, described an "insane" level of interest from AI firms, likening it to a "gold rush." Simple Closure is launching Asset Hub to facilitate the sale of code, Slack archives, and emails from companies winding down, after meticulously removing all personally identifiable information to address privacy concerns. They have facilitated nearly 100 deals, recovering over $1 million for founders, with individual companies typically receiving between $10,000 and $100,000. A competitor, Sunset, also acquires defunct company data, with pricing influenced by company size, age, and "data richness," such as the link between JIRA tickets and code commits. Certain industries, like healthcare and finance, command a premium. However, this practice raises significant privacy concerns. Mark Rotenberg of the Center for AI and Digital Policy questions whether employers should be allowed to sell internal communications to third parties, even if employees have relinquished intellectual property rights, as it goes against employee expectations regarding the use of their messages.