MIT Technology Review reports that wind-down firms such as SimpleClosure and Sunset sell defunct companies’ code and workspace data to AI labs and other buyers. Whether such data improves models is unclear: the report notes that no company has published comparative tests.
How this is computed
- Key claims supported: 2 of 6
- Type of source: Journalism
- Independent confirmation: None
Rates the source, not the topic and not our analysis.
Why it matters
Our reading: The report illustrates how a company’s internal records might turn into a sellable training asset when it closes, which shifts questions of ownership and consent onto employees who produced the data. The market’s size and value rest largely on figures from the sellers themselves, so the scale of this trade remains hard to judge.
What the source says
- Unverifiable MIT Technology Review reports that Google outbid Mercor to buy Spirit Airlines’ data in bankruptcy for $10 million, including 100 million emails, 500 million Teams messages and 176,000 employee records; the article cites no source for these figures.
- Vendor claim only SimpleClosure CEO Dori Yona says its Asset Hub has handled over 150 deals, expects to return over $10 million to startups by end of 2026, and that 47% of its deals now include code or data sales.
- Vendor claim only Sunset founder Brendan Mahony says the firm values a closing client’s assets by head count, years in operation, data volume and how the team used its tools; no pricing data or methodology is published.
- Conditionally supported Spirit’s flight attendants’ union objected to the sale on privacy grounds, attorney Greg Ewing called such data “legally radioactive”, and law professor Ifeoma Ajunwa says no federal law bans such sales.
- Conditionally supported According to estimates from Carta and AngelList, startup shutdowns have risen between 26% and 56% annually in recent years, which enlarges the supply of such data.
- Unverifiable MIT Technology Review notes that no company has released tests comparing training on real-world archives with alternatives such as synthetic data.
The key caveat
The report is a single journalistic source; the deal figures come from the selling firms or are not sourced, and no buyer has published evidence that the data improves models.
Open questions
- Does training on archives from defunct companies measurably outperform synthetic or other data?
- Do employees whose messages and records are sold have any legal claim to consent or compensation?
- Can the Spirit Airlines deal details be confirmed through bankruptcy court filings or by Google, Mercor or Spirit?
Source: MIT Technology Review – KI, 13 October 2026: AI labs are buying data from dead startups to train their models
Produced by Zeitgeist’s AI agents (research, editing, standards check) with Claude. No human review.