Spirit Airlines has not flown a plane since May. But its old internal paperwork just became one of the more interesting stories in AI this year.
Google won a bankruptcy auction for a huge trove of Spirit's internal data: emails, payroll files, spreadsheets, internal wikis, and business software code. The price was 10 million dollars. The file itself is enormous, covering roughly 3.4 million payroll records and 100 million emails across 80,000 accounts, with employee data going back to 1986.
Google was not the only bidder. Another AI training company, Micro1, says it later offered 12.5 million dollars, but only after the formal auction had already closed, so it is unclear if that offer counts. A third company, Mercor, had also bid, offering 7.5 million dollars.
None of this is really about Spirit Airlines. It is about what AI companies now consider valuable. For years, the internet's public writing, articles, books, forum posts, was the main fuel for training AI. That well is running low, and it does not teach an AI how a real business actually functions day to day.
Internal company records do. Slack messages, email chains, spreadsheets, and support tickets show how decisions get made, how disputes get handled, and how people actually talk to coworkers. Micro1 uses this kind of data to build fake companies for training AI, complete with realistic software tools, so an AI agent can practice acting like an employee before it is trusted with real work. The company says it hears from more than 200 businesses a day wanting to sell their records, and typically pays between 100,000 and 2 million dollars per deal.
This creates an odd new incentive. A struggling company's old files, the boring stuff nobody archives carefully, might now be worth more than some of its physical equipment.
It also creates a real privacy problem. A flight attendant union representing roughly 17,000 former Spirit workers has formally objected to the sale, arguing that stripping names from records is not enough. Details like which staff got disciplined, which routes generated complaints, or what people said about management can still identify someone even without a name attached.
There is no strong law stopping this. Work documents belong to the employer, not the person who wrote them, and there is little pushed regulation on what happens to that data once a company goes bankrupt or gets sold. A similar fight already played out over 23andMe's genetic data during its own bankruptcy.
For any business owner, the practical point is simple: your company's internal files are not just clutter sitting in old drives. They are now a line item that lawyers, bankruptcy trustees, and buyers will fight over. If your company ever restructures, sells a division, or shuts down, expect this exact question to come up: who gets the data, and what happens to the people whose names are in it.