AI Labs buys data from defunct startups to train its models
startup
| Source: MIT Tech Review | Original article
AI labs are purchasing data from defunct crypto startups, turning anonymous user archives into revenue, with one founder earning about $25,000 from his company's code base.
AI laboratories are increasingly purchasing data sets from defunct startups to enrich the training material for their large‑language models. The practice came to light when Keelan Miskell, who wound down his crypto‑focused venture earlier this year, discovered that the company’s digital archives could be sold. Although the startup’s users were anonymous and the firm had only operated for a few years, Miskell secured roughly $25,000 for its code base and a still‑pending portion of its data.
The transactions illustrate a growing market for “orphaned” data—software, logs and user‑generated content that would otherwise be discarded. For AI developers, such material offers a low‑cost way to expand the breadth of training inputs, potentially improving model performance on niche domains. For founders, it creates a modest revenue stream from assets that were previously assumed valueless.
The development matters because it raises fresh questions about data provenance, intellectual‑property rights and privacy. When AI models are trained on code and user data that were never explicitly licensed for such use, the risk of inadvertent infringement or exposure of sensitive information rises. Regulators are already scrutinising AI deployments; U.S. banking supervisors, for example, have begun questioning firms about the safeguards surrounding agentic AI and the origins of the data they ingest. The emergence of a secondary market for dead‑startup data could intensify that oversight.
Going forward, observers will watch for clearer industry standards on data acquisition, possible legal challenges from original creators, and any regulatory moves that require AI firms to disclose the provenance of training material. The trend also suggests that more founders may seek to monetize dormant code repositories, turning the end of a startup into a modest, data‑driven windfall.
Sources
Back to AIPULSEN