Eigen RadarAI
Analysis

Failed startups sell workplace archives for AI training

An MIT Technology Review investigation reports a growing market for the code, messages and documents left by closing startups. Buyers use real workplace records to train AI systems on tasks and decisions, while brokers say they remove personal data before a sale. Workers may still have little say over records they created, and proceeds usually flow to creditors or investors. Independent tests of the training benefit have not been published.

Artificial Intelligence··Evening
In a clearing office, a woman opens an archive box holding files and data drives.

Closing startups find buyers for their records

Founders shutting down startups are selling code and workplace archives to companies developing AI systems, according to a new MIT Technology Review investigation. One founder, Keelan Miskell, says his crypto venture’s code sold for about 25,000 dollars. A further payment of around 15,000 dollars for workspace material is still pending. Such archives can include email, chats, documents and material kept in common work tools.[1]

SimpleClosure began brokering archive sales alongside its startup closure work in April. Through Asset Hub, founders can offer code and internal records to buyers. Its reported deal count exceeds 150, and it projects distributions to startups above 10 million dollars before year-end. Those totals are company statements. They describe the reported market’s scale but do not reveal the value or data use of every transaction.[1]

Real work histories become training environments

AI training company Surge AI uses purchased work records to build practice settings for AI agents handling software fixes and legal drafts. Its researcher says records with disagreements and difficult choices are especially useful. Real archives can preserve details that synthetic examples miss. Yet the companies have not published comparative experiments establishing that training on this material works better than other methods.[1]

Another closure provider, Sunset, assesses potential archive value using a company’s size, years of operation, data volume and patterns of tool use. Buyers are therefore interested in the history of ordinary work, not only finished products. That history may include decisions, failed attempts and conversations never written for public release. The investigation finds a commercial path from a company’s closure to model training, while the resulting model benefit remains unmeasured publicly.[1]

Employee records raise consent questions

Companies involved in the sales describe removing identifying information. Attorney Greg Ewing points out that email and calendars may still expose health visits and family pickup arrangements. Ifeoma Ajunwa, a law professor at Emory University, says workers should explicitly permit sales of records they helped create. Their concern is about the people represented in the archive, beyond the founder signing the transaction.[1]

Sale proceeds generally go to senior creditors and investors, according to the investigation. Founders may also have limited visibility into who eventually buys the material and how it will be used. The reported transactions show why closing a startup does not end the commercial life of its records. Questions about employee consent and downstream use remain alongside the payments and the claimed training value.[1]

References

  1. News sourceMIT Technology ReviewFailed startups’ workplace archives are being sold for AI training↩1↩2↩3↩4↩5↩6