AI companies are paying for the code and files of startups that are shutting down, to train their models. One founder closing his company received around $25,000 for the codebase alone. Such data is valued because it is real and not publicly available. The practice raises questions about the privacy of the former customers.
What it means: Buying data from companies that are shutting down shows how much the value of a startup's code and records has changed: even when the product fails, the source code, documentation and internal files are real, non-public material that AI labs use to improve their models on genuine programming problems. Around 25,000 dollars for one company's code is not a fortune, but for a founder winding down it can cover liquidation costs, which creates a strong incentive to sell as many files as possible. That is exactly where the risk lies, because if those files include customer details, support messages or logs, selling them may breach the terms of service or GDPR, since users handed over their data for a different purpose. For any founder considering such a sale, the right first step is a legal review of what is allowed and a clean separation of code from personal data, and for customers of small vendors it is a reminder to check what happens to their data if the provider stops operating.
Read the original on MIT Technology Review
Related service: Custom software