AI Labs Buy Workplace Data to Train Enterprise Agents

Artificial intelligence developers are paying for corporate emails, Slack messages, source code, and internal documents to train AI agents that perform office tasks. The market, which includes archives from defunct startups and data licensed by operating businesses, is expanding as technology companies seek detailed records of how employees make decisions and complete work, according to reporting by MIT Technology Review, Business Insider, and KuCoin.

SimpleClosure, a company that helps startups shut down their operations, added data sales to its services in April. Its Asset Hub platform has handled more than 150 deals and expects to return over $10 million to startups by the end of 2026, according to MIT Technology Review. Chief executive Dori Yona said 47% of the company's deals now include code or data sales.

Keelan Miskell, who wound down a cryptocurrency startup, received approximately $25,000 for its codebase and was awaiting another payment likely to reach $15,000 for workspace data stored in Google, Notion, Figma, and Slack. His experience illustrates how digital records can become financial assets after a business closes.

Demand extends beyond failed companies. William Almeida, chief revenue officer at AI training startup Micro1, said the company announced plans earlier in 2026 to pay medium-sized businesses between $100,000 and $2 million for access to anonymized data. The materials can include records showing how teams coordinate projects, develop products, and make decisions.

Dan Novaes, CEO of Mode.Inc, described the market's growth in comments reported by KuCoin.

"The current demand is astronomical," Novaes said.

According to Novaes, dataset value depends partly on employee head count, while exclusivity and the amount of available information also influence pricing. Demand is particularly strong for records related to legal and accounting work, where AI developers are seeking to automate professional tasks.

Surge AI also acquires data to create simulated digital workplaces where AI agents can practice complex tasks. Nick Heiner, who leads the company's work on these environments, said authentic records can help reproduce the varied decisions and disagreements that occur in real organizations.

Heiner said that a company's Slack history can be more valuable than a larger collection of emails, messages, and calendars if it contains a wider range of meaningful situations. However, MIT Technology Review reported that no company had publicly released comparative tests establishing how much better models perform when trained on real workplace records rather than synthetic alternatives.

Some transactions have involved substantially larger datasets. During Spirit Airlines' bankruptcy in August 2026, Google outbid AI data company Mercor to acquire decades of corporate records for $10 million. The collection included 100 million emails, 500 million Microsoft Teams messages, and 176,000 employee records, according to MIT Technology Review.

Doug Kreuzkamp, CEO of Springshot, an airline services company that had worked with Spirit Airlines, said he learned about the purchase through media coverage.

"I found out—less than 48 hours ago, and from the media—that the intellectual property we spent 15 years building might be handed over," Kreuzkamp told Business Insider.

Springshot had helped coordinate work among flight crews, gate agents, and cleaners. The transaction raised questions about whether employees and business partners understood how their information could be used after Spirit's collapse.

Privacy concerns have also prompted opposition from Spirit's flight attendants' union. According to KuCoin, Senator Elizabeth Warren and more than 100 members of Congress wrote to Google and Spirit the previous week, urging protections for the privacy of thousands of people whose information could be affected by the sale.

Companies involved in these transactions say they remove sensitive information before transferring datasets. Suppliers may use data scientists to eliminate personally identifiable information and exclude human resources records, while buyers may conduct additional screening. However, private details can remain embedded in calendars, emails, and other business communications.

Ifeoma Ajunwa, an Emory University law professor and author of The Quantified Worker, said the central question is whether employees' original permission to collect their information also covers its subsequent sale.

"The issue is whether the consent employees gave when their data was collected truly extends to the sale of that data. I think this sits in an ethical gray area," Ajunwa said.

Ajunwa also argued that employers should obtain explicit consent before selling employee data. MIT Technology Review reported that the United States has no federal law specifically prohibiting such sales, leaving companies with limited federal privacy restrictions in this area.

The financial benefits may also bypass the people whose work generated the records. Proceeds from asset sales typically go to senior creditors and investors rather than former employees, while business owners may have limited information about buyers, intended uses, or whether offers reflect fair market value.

Grant Gooley, whose company MyPortal Marketing licensed materials to Micro1, said the company supplied Figma and Canva files, customer onboarding checklists, and Google Ads data without selling client-specific materials.

"If AI is destined to be trained, why not participate—and do so responsibly?" Gooley said.

Meanwhile, MIT Technology Review reported that Google's purchase of Spirit's records was part of a wider expansion in demand for corporate information, with SpaceX's AI unit weighing purchases from struggling startups. As companies pursue more capable workplace agents, the market continues to raise questions about data ownership, employee consent, cybersecurity, and the distribution of proceeds.