Privacy by Editorial Team

AI labs are buying bankrupt startups' data to train their models

From Spirit Airlines' archive to a small crypto firm's Slack, data from failed companies is being resold to AI labs, with few legal limits.

Spirit Airlines declared bankruptcy in spring 2026, per Wired. In August, as part of that bankruptcy, Google paid $10 million for the carrier’s decades of data, including 100 million emails, 500 million Teams messages, and 176,000 employee records. The flight attendants’ union objected to the sale on privacy grounds, per MIT Technology Review; Wired reported that the Association of Flight Attendants filed the objection in bankruptcy court days after Google’s winning bid was announced. Google had outbid the AI data company Mercor for the package.

The Spirit deal is the largest in MIT Technology Review’s reporting, but the same pattern is being built out by a cottage industry of startup-liquidation firms. When a venture-backed company winds down, its code, emails, support logs, and chat histories do not disappear. They are packaged, priced, and resold - and the buyers include foundation AI labs, vendors that build simulated workspaces for AI training, and tech companies building their own models.

A new asset class

The clearest window into this market is SimpleClosure, a startup-liquidation services firm that added data sales to its offering in April 2026 through a marketplace it calls AssetHub. The CEO, Dori Yona, told Upstarts Media on September 3, 2026 that more than 350 companies had listed assets on the platform by that date, with roughly 500 million lines of code and around 3,000 repositories up for grabs. SimpleClosure expects to return more than $10 million to listing startups by the end of 2026, and 47% of its deals now include code or data sales, per MIT Technology Review.

SimpleClosure is not alone. A rival called Sunset, run by founder and CEO Brendan Mahony, also has a rapidly growing data resale business. MIT Technology Review reports that Sunset values a dying client’s assets by head count, years in operation, volume of data, and how the team used its tools. Mahony frames the goods as “operational exhaust”: “This operational exhaust data is like the wood shavings that everyone thought had no value, but it actually has a ton.” Surge AI, which builds “reinforcement-learning gyms” for AI agents, also buys in, with Nick Heiner leading the environments work there, per the same MIT Technology Review report.

Supply is growing too. Carta and AngelList estimates cited by MIT Technology Review show shutdowns rising between 26% and 56% annually in recent years. Each shut-down company is a potential listing.

What actually gets sold

The data leaving these deals is not just code. Workspace exports from failed startups include emails, customer records, support logs, old Slack channels, Jira tickets, Figma files, and Notion documents. Upstarts Media reports that the premium listings are databases of rare provenance, such as 100 million images of eyeballs or a billion interactions with service reps.

Founder Keelan Miskell, who wound down his own crypto startup this year, netted roughly $25,000 for the codebase and has a pending amount, likely around $15,000, for workspace data in Google, Notion, Figma, and Slack. He told MIT Technology Review: “At least all that stuff we were doing for two and a half years is worth something to someone somewhere.”

Suppliers employ data scientists to scrub personally identifiable information and exclude HR records, and buyers say they scrub too, per MIT Technology Review. MIT Technology Review notes that the founders selling the data have “little to no visibility into who is buying the data and what they plan to do with it.”

Two voices in the MIT Technology Review piece frame the gap. Greg Ewing, an attorney who advises AI startups, called the data “legally radioactive” and described a “Wild West of AI training” that he says is only starting to give way to proper licensing agreements and vetted use. Ifeoma Ajunwa, a law professor at Emory University and the author of The Quantified Worker, was more direct: “Employers should have express consent before selling workers’ data.”

She adds that there is no federal law banning such sales. Ajunwa also has a phrase for the resulting economics: proceeds typically flow to senior creditors and investors, not employees, which she describes as a form of “captured capital.”

Demand is still widening. Bloomberg reported in September that SpaceX’s AI unit is weighing buying data from struggling startups, per MIT Technology Review. SimpleClosure says the number of prospective buyers on its marketplace grew ninefold after April, with AI labs and data labeling shops among them, per Upstarts Media.

Why This Matters

With no federal law banning such sales, a worker whose employer just shut down has few obvious ways to stop their old Slack messages or emails from being sold as training data. The firm that brokered the deal (SimpleClosure, Sunset, or someone similar) and the buyer (Google, a foundation model lab, a data-labeling shop) both say they scrub personal data, and Ewing’s point is that emails and calendars still hold details like doctor appointments, school pickup arrangements, and date-night plans.

For startups still operating, this is now a near-term privacy question, not a hypothetical one. Every line written into a company Slack, every Jira ticket, every recorded customer-support call is potentially a future listing.

The Bottom Line

A loose network of shutdown-services firms has built a secondary market in dead startups’ data, and foundation AI labs are among the buyers. The biggest deal in MIT Technology Review’s reporting is Google’s $10 million purchase of Spirit Airlines’ data in August as part of Spirit’s bankruptcy, in which 100 million emails and 500 million Teams messages went to a single buyer. Ajunwa notes there is no federal law banning such sales, and even the founders selling the data have little to no visibility into who is buying it and what they plan to do with it.