Bankruptcy court filings show Spirit Airlines selected Google LLC as the winning bidder for its deidentified corporate data on Friday, August 14, 2026, at a price of ten million dollars. Mercor.io Corporation, a startup that supplies human-labelled data to artificial intelligence developers, placed the backup bid at seven and a half million. A hearing to approve the transfer is set for Wednesday, August 19.

The notice was filed with the United States Bankruptcy Court for the Southern District of New York at 22:00:10 on August 14, docketed as entry 1463 in Case No. 25-11897 (SHL). Spirit Aviation Holdings, Inc. and five affiliated debtors entered Chapter 11 on Friday, August 29, 2025. The court approved bidding procedures on June 22, 2026, authorising sales of the debtors' assets free and clear of liens, claims and other interests. Under those procedures the debtors ran a virtual auction for a lot the filings call the Deidentified Data.

According to the notice, the debtors consulted the Consultation Parties before selecting Google's bid as the highest and best. The two bill-of-sale drafts attached as Exhibit A and Exhibit B are close to identical in structure. Both are section 363 sales. Both are governed by New York law. Both waive jury trial. The purchase price is the principal difference, along with one clause on how the files physically move.

What is actually being sold

The asset schedule attached to both agreements runs to a single dense page and splits Spirit's data estate into rows marked Included or Not Included. The column that carries those designations is headed "Google's Data Purchase Request", and the same schedule, with the same header, is attached to the Mercor alternate agreement.

The included side is corporate rather than consumer. Under productivity systems, the schedule lists 100,000,000 emails across 80,000 Microsoft 365 accounts dating from 2018, alongside 17,082,644 OneDrive items, 20,577,677 SharePoint items and 500,000,000 Teams records. Information technology tickets held in ServiceNow number 667,563.

Software is a separate line. The schedule records 516 source code repositories covering web, mobile and backend systems, holding roughly 30 million lines of code, with 43,170 pull requests including descriptions, comments and review threads, and 372,585 commits carrying full git history, diffs, author information and timestamps. Issue trackers, continuous integration logs, automated test results and code coverage reports are included. Exhibit A extends the software grant to source code, object code, architecture, software models, plugins, algorithms, libraries, compilers, subroutines, tools and application programming interfaces, together with technical specifications, user manuals and training materials.

Commercial operations carry the largest raw counts. Revenue data held in Navitaire runs from May 2008 and covers 190,312,864 passenger name records and 7,510,221,520 transactions. Competitor fare observations sourced from Infare since January 2021 total 7,250,630,887 flight observations. Booking curve records from Navitaire and Kambr since January 2023 hold 3,530,769,431 farebasis observations. Irregular operations records held in ITS since February 2018 count 3,000,347,472 reaccommodations. Base fare records from PROS number 24,510,638 since July 2025. Payment processing through Elavon contributes 78,432,834 transactions since December 2022, and onboard purchasing through Retail in Motion adds 29,000,147 since October 2021. Refunds, credit shells and vouchers add 9,292,480, 18,244,428 and 12,139,319 records respectively, the voucher series reaching back to March 2009.

Operational and maintenance data covers 763,391 Spirit flights flown since April 2023, 74,163,945 industry flown-statistics records since January 2017, 57,362,400 scheduled flights, a crew base of 4,600, and 5,014,676 crew pairings since January 2021. Maintenance systems held in TRAX record 787,452 parts received, 459,828 scheduled tasks and 313,646 parts transferred. Fuel records from SkyMetrix count 1,239,196 fuel slips since June 2021.

Human resources data is included in full. Employee records held in UKG date from August 1986 and cover 175,658 employees, with 3,426,618 payroll records and 148,018 employee tax forms since June 2016, plus 1,092,000 time card entries in Kronos since December 2012. Corporate and crew training records, recruiting documents from iCIMS and business travel records are all marked included.

Finance and legal round out the set: 97,603 vendor identifiers in SAP, 851,590 invoices in Coupa since January 2019, 1,849,736 cost accounting records, financial models and forecasting files, board presentations covering capital projects and budget walkthroughs, corporate development materials including deal pipelines and merger fairness opinions, and a legal category covering forms, memos, litigation case files, and redlined contract histories spanning commercial agreements, merger and acquisition contracts and employment agreements.

What is being withheld

Every customer-facing row is marked Not Included. That covers 97,500,000 customer profiles held in Navitaire since May 2017, 50,200,000 Free Spirit loyalty members, 2,200,000 Savers Club members, 740,000 credit card holders and 44,000,000 earned miles. Marketing systems are excluded too: 13,700,000 active email addresses in Oracle Responsys, 2,700 year-to-date campaigns, social posts in Sprinklr since March 2022, and survey data in Qualtrics covering 1,446,033 post-purchase responses.

Contact centre material is out. That means 15,784,473 handled chat sessions in Quiq since January 2024, 30,865,471 call recordings in Calabrio and Cresta since January 2021, and 7,341,857 telephone numbers. Website analytics held in GA4 and Databricks since July 2024, covering 43,117,864 unique users, 41,414,953 searches and 2,418,201 purchases year to date, is excluded. So is the regulatory complaint file: 2,491,715 disability service requests, 338,531 Department of Transportation complaints logged in Netracer since March 2025, and 7,790 logged in Navitaire since July 2023.

A tension sits inside the drafting. Exhibit A of both agreements describes the assets as including workflow and process data, "including, without limitation, marketing data (e.g., campaigns, content and demand generation data)". The schedule attached to that same exhibit marks the email behaviour, campaign and social rows Not Included. The narrower schedule controls, since Exhibit A defers to it explicitly, but the general grant language and the row-level designations pull in opposite directions.

The deidentification machinery

Section 1(a)(i) of both agreements states the parties' express intention that the assets exclude anything that relates to, describes, or is reasonably capable of being associated with a consumer, or that counts as personal data under applicable law. The definition then carves out an exception for Deidentified Data.

That produces an unusual sequence. The agreement asserts the assets contain no personal data, then requires a process to remove personal data from them before delivery. Under Section 1(a)(ii), Spirit must first deliver the files to one or more third parties acceptable to or designated by the buyer, called a Deidentification Agent, which takes reasonable measures to remove or transform data elements so the result cannot be associated with, used to infer information about, or otherwise linked to a particular consumer. Only then does the material move to the buyer.

The certification standard is written into Section 3(c). The agent must certify to the buyer's reasonable satisfaction that the assets were deidentified in accordance with applicable law, including the standard set out in the California Consumer Privacy Act "regardless of whether such statute does, or does not, apply to the Assets", and, for protected health information or other consumer health data, the standard at 45 C.F.R. section 164.514. The clause adds a requirement that cuts against aggressive transformation: the process must run "while preserving referential integrity across the data set". A dataset that keeps referential integrity keeps the joins between records intact, which is what makes it useful for training and analysis, and which is also what makes linkage attacks feasible.

The buyer pays for the process. Section 1(a)(iii) states that deidentification costs sit with the buyer and do not reduce the purchase price. Section 4 gives the buyer a right to review and comment on the process, whether run by Spirit or by an agent, with Spirit obliged to give good faith consideration to those comments.

A public commitment written into a private contract

Section 1(c) is the clause with the widest reach beyond the parties. The buyer publicly commits to maintain and use the Deidentified Data in deidentified form and not to intentionally associate it with any person or household. If the buyer has not made that commitment in a broadly accessible public document, such as a posted privacy notice, the agreement itself is deemed to constitute the public commitment required under applicable data protection law.

The same section then permits onward transfer. The buyer may pass Deidentified Data to third parties provided it contractually obligates those transferees to comply with Section 1(c). Nothing in the drafting caps the number of hops or requires disclosure of who receives the data.

Spirit's own retained rights are narrower on paper and broader in one specific respect. Section 1(b)(ii) lets Spirit keep copies solely for winding down the business or complying with law, and bars any sale or transfer of the assets to anyone other than the buyer. The proviso that follows carves an exit: Spirit "may sell its customer data list, which includes individual traveler spend aggregated by year, to third-parties in hospitality or travel industries." The passenger list, in other words, is not being retired. It is being routed to a different buyer pool.

Two bidders, one procedural difference

The Mercor agreement carries an Exhibit C the Google agreement does not. It specifies that Spirit personnel or its third-party vendors transmit complete data exports, after deidentification, to the buyer through a secure upload facility at data.mercor.com, in formats native to each source system. Mercor's Section 4 correspondingly obliges Spirit to assist with "Buyer's completion of the data extraction processes set forth on Exhibit C". Google's version omits both. Its delivery specification instead appears in the asset schedule, which asks that Microsoft 365 material stay inside the native Microsoft 365 environment during transfer, that non-Microsoft systems such as SAP, Navitaire and UKG export to CSV, JSON or native SQL dumps, that repositories arrive as bare git repositories or bundle files with one archive per repository, and that commit, pull request and issue metadata arrive as JSONL.

Counsel differs as well. Davis Polk and Wardwell LLP acts for the debtors, with Marshall S. Huebner, Darren S. Klein and Christopher S. Robertson named. Cleary Gottlieb Steen and Hamilton LLP represents Google, with Sean O'Neal, Glenn McGrory and Charles Allen listed. Orrick, Herrington and Sutcliffe LLP represents Mercor, with Mark W. Seneca, John Harrison and Evan C. Hollander named, and Arek Wardak listed as Mercor's general counsel. A confidentiality agreement between Google and Spirit Aviation Holdings is dated June 18, 2026, four days before the bidding procedures order.

Both agreements disclaim every warranty in capitalised text. The buyer takes the assets "AS IS" and "WHERE IS" with all faults, waiving claims about merchantability, fitness, non-infringement and the accuracy or completeness of anything disclosed. Spirit warrants only that it owns the assets and has authority to sell them, and those warranties do not survive closing. Claims based on fraud are preserved.

The anonymisation argument, running in both directions

Alan Chapell, a privacy attorney who serves as chairman of the board of the Network Advertising Initiative and hosts The Monopoly Report podcast for Marketecture Media, posted on LinkedIn that the transaction sits awkwardly against Google's litigation positions.

"In a NY bankruptcy court, Google is about to purchase a trove of data from the Spirit Airlines bankruptcy. But don't worry they say, it's not personal data - it's been anonymized," Chapell wrote. He continued: "In two separate antitrust/competition venues, Google is OBJECTING to the provision of search data to rivals - effectively stating that it's all but impossible to truly anonymize data."

Chapell added that the company "has assured us all that their use of privacy enhancing technologies will effectively protect privacy, eliminate the possibility of re-identification and make advertising better", and closed: "It's kind of amazing how good Google is at protecting data - and apparently how bad the rest of the world is...."

The venues he refers to are documented. Judge Amit Mehta ordered Google to share its Glue query-interaction systemand RankEmbed training data with qualified competitors in September 2025. Google appealed the resulting six-year judgment in January 2026 and asked the D.C. Circuit in May 2026 to throw the liability finding out entirely. In Brussels, the European Commission adopted binding Digital Markets Act decisions in July 2026 requiring anonymised query, click, ranking and view data to flow to rival engines, after finding that the company's first compliance offer had stripped out between 90 and 100 percent of unique search queries. The technical specification behind that remedy ran to 29 pages of field-level anonymisation requirements. Google did not comment publicly on the Spirit purchase in the documents reviewed.

Reaction: a price argument, not a privacy one

Matt Mickiewicz, founder of SlashSpend.ai and co-founder of Hired.com and 99designs, surfaced the filing on X at 19:34 on Monday, August 17, 2026. "AI Labs are so desperate for fresh training data, that even the emails & documents of bankrupt Spirit Airlines get auctioned off for $10 million," he wrote. The post recorded 5,609 views.

The thread beneath it argued about valuation rather than consent. Trent Krupp replied: "This data is actually super interesting. The number of operating years, the industry and the completeness of the data would be unique." He added: "This was actually a steal I think. Street value for this might be 3-5x what GOOG payed."

Mickiewicz asked why rivals had stayed out: "Then why didn't OpenAI or Anthropic bid $30m+". Krupp's answer described a resale model: "You sell the data multiple times. Mercor should have bid over $10m to win it, and then sell it 3-5 times to frontier labs." Mickiewicz later summarised the episode in four words: "new asset class unlocked."

Why this matters for marketers

Three things follow for people who buy media, run data operations, or write privacy notices.

The first is definitional. European regulators moved the anonymity goalposts six weeks before this auction. The European Data Protection Board replaced the 2014 Article 29 Working Party test with an updated framework adopted on July 7, 2026, adding an explicit No Inference criterion that fails where a relevant entity could draw a specific and meaningful conclusion about an individual without isolating a record or linking it externally. The Spirit agreements certify against the California standard and the federal health rule, not the European one, and the requirement to preserve referential integrity is precisely the property the inference criterion targets. Membership inference, under the European framework, can breach the criterion on its own. The same plenary adopted guidelines concluding that consent will most probably not work as a legal basis for training-scale collection, and finalised blockchain guidance in the same session.

The second is that insolvency has become a data acquisition channel. The California attorney general sued 23andMe in May 2026 over genetic profiles while separately opposing the sale of Californians' genetic information in that company's Chapter 11 proceeding, following a UK penalty of 2.31 million pounds whose enforcement was suspended pending the bankruptcy outcome. Spirit's structure is different in kind: the consumer records stay out of the lot and the corporate exhaust goes in. But the mechanism is the same. A section 363 sale transfers assets free and clear, on a timetable measured in days, with a three-day objection window. The distinction between a consumer database and a corporate one is a drafting choice made by the seller, not a statutory boundary. Hughes Satellite Systems filed its own Chapter 11 on August 2, 2026, and the same asset category exists in every distressed technology and media estate.

The third is pricing. Reddit's licensing arrangement with Google, reported at roughly 60 million dollars a year, sets one benchmark for conversational text. The New York Times chief executive has put the cost of producing the company's journalism at close to 2 billion dollars a year as a floor for licensing talks, and Reddit's leadership has claimed the platform supplied roughly a third of one major model's training corpus while booking 36 million dollars of licensing revenue in a quarter. Against those figures, ten million dollars for an operating airline's full internal record, including two decades of pricing behaviour and the source code that produced it, is a low number. Every organisation holding a comparable corpus now has a public comparable, and it is a distressed one.

There is also an employment dimension that has not been litigated. The included set covers 100 million emails from 80,000 accounts, 175,658 employee records reaching back to 1986, payroll and tax forms, training records and recruiting documents. Research published in April 2026 found that workplace applications collect far more data than employees expect, with the average tool gathering close to 20 distinct data types. The deidentification clause in the Spirit agreements addresses consumers. Its operative language is about anything "reasonably capable of being associated with a consumer". Employees are not consumers of their employer's mailbox.

Objections were due by 16:00 Eastern on Monday, August 17. Parties wishing to appear had to register by 11:00 on Tuesday, August 18. Judge Sean H. Lane hears the matter over Zoom for Government at 11:00 on Wednesday, August 19. All objecting parties are required to attend, and failure to appear may result in relief being granted by default.

Timeline

  • August 1986 - Earliest employee records in Spirit's UKG human resources system, covering 175,658 employees in total
  • May 2008 - Earliest revenue records in Navitaire, the system holding 190,312,864 passenger name records and 7,510,221,520 transactions
  • August 29, 2025 - Spirit Aviation Holdings, Inc. and five affiliates file voluntary Chapter 11 petitions in the Southern District of New York
  • September 2, 2025 - Judge Amit Mehta orders Google to share Glue query-interaction data and RankEmbed training data with qualified search competitors
  • January 16, 2026 - Google files notice of appeal against the six-year behavioural remedies
  • April 2026 - The European Commission's 29-page anonymisation specification for Digital Markets Act search data sharing becomes public
  • May 22, 2026 - Google asks the D.C. Circuit to reverse the search monopoly finding entirely
  • June 18, 2026 - Confidentiality agreement executed between Google LLC and Spirit Aviation Holdings, Inc.
  • June 22, 2026 - Bankruptcy court enters the Bidding Procedures Order, docket entry 1213
  • July 7, 2026 - The European Data Protection Board adopts Guidelines 02/2026 on Anonymisation, adding a No Inference criterion, alongside web scraping guidelines for generative AI
  • July 2026 - The European Commission adopts binding decisions requiring Google to share anonymised search data with rivals
  • August 14, 2026 - Virtual auction for the Deidentified Data; Google LLC selected as successful bidder at ten million dollars, Mercor.io Corporation as alternate at seven and a half million; notice filed at 22:00:10 as docket entry 1463
  • August 17, 2026 at 16:00 Eastern - Objection deadline
  • August 17, 2026 at 19:34 - Matt Mickiewicz posts the auction results to X, drawing 5,609 views
  • August 18, 2026 at 11:00 Eastern - Deadline for parties to register an electronic appearance
  • August 19, 2026 at 11:00 Eastern - Sale hearing before Judge Sean H. Lane, conducted over Zoom for Government

Summary

Who: Google LLC was selected as the successful bidder by Spirit Airlines, LLC and its affiliated debtors in possession, with Mercor.io Corporation named as the alternate bidder. Davis Polk and Wardwell LLP represents the debtors, Cleary Gottlieb Steen and Hamilton LLP represents Google, and Orrick, Herrington and Sutcliffe LLP represents Mercor. Judge Sean H. Lane presides. Alan Chapell, chairman of the Network Advertising Initiative board, and Matt Mickiewicz, founder of SlashSpend.ai, commented publicly.

What: A section 363 sale of Spirit's deidentified corporate data for ten million dollars, covering 100 million emails, 516 source code repositories holding roughly 30 million lines of code, 7,510,221,520 revenue transactions, 175,658 employee records and finance, legal and operations archives. Customer profiles, loyalty membership, contact centre recordings, marketing email lists and regulatory complaint files are excluded. Spirit retains the right to sell its customer list separately to hospitality and travel buyers.

When: The auction ran on Friday, August 14, 2026, with the results notice filed at 22:00:10 the same day. Objections were due by 16:00 Eastern on August 17, appearance registration closed at 11:00 on August 18, and the sale hearing is scheduled for 11:00 Eastern on Wednesday, August 19, 2026.

Where: The United States Bankruptcy Court for the Southern District of New York, Case No. 25-11897 (SHL), with the hearing conducted over Zoom for Government. Spirit's mailing address is 1731 Radiant Drive, Dania Beach, Florida.

Why: Chapter 11 asset sales convert data holdings into cash for creditors on a compressed timetable. Demand from artificial intelligence developers for large, coherent, non-public corpora has given corporate records a market price they did not previously carry. The transaction establishes a public benchmark for that price and, because the deidentification standard is certified against California and United States health rules rather than the European framework adopted six weeks earlier, it leaves open how the same dataset would be classified under the criteria European regulators now apply.