When Spirit Airlines shut down in May 2026, its bankruptcy estate held an unusual asset: a corporate archive of about 100 million emails and 500 million Microsoft Teams messages. Google won the auction for a de-identified copy with a $10 million bid. Unions for Spirit’s flight attendants and pilots are fighting the sale, and buyers hunting for the same kind of data are now making offers to companies that are still in business.
Dead Startups Opened the Market
Demand for workplace data appeared suddenly last year, MIT Technology Review reported in October, when AI labs reached the limits of public and synthetic datasets. SimpleClosure, a firm that helps failing startups wind down, added data sales in April. Its Asset Hub platform has handled more than 150 deals, and code or data sales now figure in 47% of the firm’s deals.
Payouts to founders have stayed modest. According to Fortune, SimpleClosure puts typical payments at $10,000 to $100,000 per company. Cielo24 CEO Shanna Johnson told Fast Company she received “hundreds of thousands of dollars” for the shuttered company’s Slack messages, emails and Jira tickets.
Spirit’s auction played at a different scale. Google won on August 14 with its $10 million bid, ahead of an alternate $7.5 million offer from Mercor, an AI data company. Micro1, another AI training-data firm, later filed a $12.5 million offer and proposed to handle the de-identification itself. The court’s privacy ombudsman said she would need more time to review that plan.
Operating Companies Are Getting the Calls
The buyers have moved past bankruptcy court. Business Insider reported on October 13 that brokers are offering operating companies six-figure sums for Slack messages, emails and CRM records, sometimes with 48 hours to accept. Micro1 announced on October 9 that it will spend $1 billion over 12 months to acquire and license enterprise operational data, and Fortune reported that its website advertises payments from $100,000 to more than $2 million for approved data packages.
Buyers want the mess. According to MIT Technology Review, real archives avoid errors that synthetic data produces, such as references to dates that do not exist. Nick Heiner, who leads environments work at Surge AI, seeks archives that capture disagreements and judgment calls. Surge AI uses them to build simulated workplaces where AI agents practice tasks such as debugging IT architecture or drafting appellate briefs.
The Price Rests on an Untested Bet
MIT Technology Review found that no company has published tests comparing training on real-world data with the alternatives, so the size of the performance gain remains unclear. Buyers quote widely different prices because each figure rests on the buyer’s own research thesis about how many useful tasks an archive can yield.
My take is that a seller handed a 48-hour deadline is pricing an asset that even the buyers cannot yet value with evidence. A short deadline argues for slowing the process down, not accepting the first number.
The Seller Carries the Legal Risk
Holding data does not give a company the right to sell it, according to Daniel Goldberg, a partner at the law firm Frankfurt Kurnit Klein & Selz, whose guidance appeared on October 1. Email and chat often contain other companies’ confidential information covered by nondisclosure agreements. A sale can waive privilege over legal communications, and Slack’s API terms bar third-party apps from using API data to train large language models. Under the California Consumer Privacy Act, selling employees’ personal information triggers notice, opt-out and risk assessment duties unless the data is properly de-identified. The US National Labor Relations Act may also protect employee messages about pay and working conditions, even in workplaces without a union.
Spirit shows the third-party problem in practice. Software vendor Springshot says the dataset includes its intellectual property, according to Skift. Greg Ewing, an attorney who advises AI startups, called this kind of data “legally radioactive” in MIT Technology Review’s reporting. A seller cannot assume the deal stays confidential either: California’s AB 2013 and the EU AI Act require generative AI developers to disclose high-level information about their training data sources, including whether they purchased or licensed the data.
De-Identification Answers a Narrower Question Than It Seems
Brokers and buyers say they strip identifying details before data changes hands. Google says it is not seeking personal information and that an independent third party will de-identify the data before it receives anything. Goldberg’s guidance points out that removing identifiers from millions of unstructured emails and chats is much harder than cleaning structured fields. It also notes that de-identification addresses whether a record links to a person, not whether its content is confidential, so a pay complaint or disciplinary note stays sensitive without a name attached.
Spirit shows the gap. The court-appointed consumer privacy ombudsman, Lucy Thomson, recommended approval on October 5 after Spirit and Google took passenger databases out of the sale, but her review did not assess risks to employees. The Association of Flight Attendants-CWA argues that links preserved across datasets could still let individuals be re-identified. In October, 121 members of the US Congress, led by Senator Elizabeth Warren and Representative Steven Horsford, asked Google and Spirit to hold off on the transaction until further safeguards protect Spirit employees’ personal data.
Employees typically see none of the money. MIT Technology Review reported that sale proceeds usually go to senior creditors and investors, and Emory University law professor Ifeoma Ajunwa told the publication that no US federal law bans such sales.
What to Settle Before an Offer Arrives
The implication for operators is that the decision on whether to sell anything should rest with a named executive and legal counsel before the first offer lands, because a 48-hour deadline leaves no time to build that process. Goldberg’s guidance outlines the groundwork: inventory which systems hold which records and over what date ranges, review customer, partner and vendor agreements, and exclude privileged material, HR and payroll records, health information, union communications and regulated data rather than relying on de-identification alone. Legal, privacy, security, HR and communications teams should review any sample before the company shares it.
Software vendors deserve the same scrutiny. Companies should check whether their vendors’ contracts allow the sale or transfer of customer data if the vendor changes hands or goes bankrupt, because a company’s archive can reach a buyer through a vendor’s estate rather than its own. Negotiating that language is easier before a vendor receives its own offer.
Spirit Is the First Test
As of early November, the Spirit sale still awaits approval from Judge Sean Lane of the US Bankruptcy Court for the Southern District of New York, after repeated adjournments since August. Whatever he decides, the conditions he sets for employee data could become the template for the next estate and the next broker. Companies that are still operating can write their own terms first, while the decision is still theirs to make.
The post Google Won a $10 Million Auction for a Bankrupt Airline’s Emails, Yours May Be Next appeared first on DataFLOQ.
