AI labs are buying data from dead startups to train their models

As large language models and other AI systems grow increasingly dependent on diverse training data, labs are turning to an unconventional source: the digital remnants of shut-down companies. According to MIT Technology Review, founders of defunct startups are selling their firms' codebases, user data, and other digital archives to AI developers seeking fresh material for model training.

The trend highlights a growing tension in the AI data economy. While major labs have long relied on web-scraped content and licensed datasets, the sheer volume of proprietary data locked inside dead companies represents a largely untapped reservoir. For founders winding down operations, selling this data offers a way to monetize assets that would otherwise be discarded.

This practice raises questions about data provenance and consent. Much of the data from defunct startups may have been collected under privacy policies that did not anticipate eventual sale to AI companies. As AI labs expand their data acquisition strategies, the legal and ethical boundaries of training data sourcing are likely to come under greater scrutiny.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.