Google Bids $10 Million for Spirit Airlines’ Internal Data to Train AI — Privacy Questions Follow

What Google Says It Is Buying

Court filings and subsequent reporting describe a dataset that includes roughly 100 million emails and about 500 million Microsoft Teams messages, along with calendars, spreadsheets, OneDrive and SharePoint files, and hundreds of code repositories containing tens of millions of lines of custom software. The package also covers operational and commercial records: aircraft operations, revenue management, crew and productivity data, marketing materials, financial databases, IT tickets, and customer-service workflows.

Google has been explicit about what it says it is not buying. A company spokesperson said Google is acquiring “the company’s internal data and custom software, but we are not buying their customer or credit card information.” In another statement: “We acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models. We will not receive any personal information from this dataset. Any data we receive will be rigorously scrubbed of any personally identifiable information by a third party before receipt.”

Passenger profiles and Free Spirit loyalty-program records are excluded from this sale. Spirit’s estate has separately reserved the ability to market certain customer-related lists to hospitality or travel buyers.

Why This Data Is Valuable for AI

Public web text is abundant. Detailed records of how people inside a company actually work are not. Internal email and chat logs, booking and pricing processes, exception handling, and operational software show decision-making, coordination, and error correction in context. That is the kind of material labs increasingly want for training and evaluating agents that are supposed to operate inside business systems rather than only generate text.

Mercor, the backup bidder, framed the same point after the auction. “Companies are sitting on decades of records that show how real work gets done, and that data is now some of the most valuable material for training and evaluating AI,” a Mercor spokesperson said.

The auction itself illustrated the demand. Google opened around $5 million. Mercor countered and later offered more if it could receive the raw data first and anonymize it itself. Google raised its bid to $10 million and won. Mercor remains the backup buyer if the Google sale is not approved.

De-Identification Is the Legal Center of the Deal

The sale is structured around de-identification. A third party is to strip personally identifiable information before Google takes possession. Court descriptions refer to “reasonable measures” so that records cannot be linked to individuals, and Google has said it will not attempt to re-identify users.

That framework is standard language. It is not a complete answer to every privacy risk. Emails and workplace chats often contain names, roles, incident details, medical or disciplinary context, and enough surrounding facts that identities can sometimes be inferred even after obvious identifiers are removed. The Association of Flight Attendants-CWA has argued that because the sale agreement requires links across datasets to be preserved — so the data remains useful as a connected enterprise record — information about individuals or small groups could potentially be reconstructed.

Sara Nelson, international president of the AFA-CWA, which continues to represent more than 5,500 of Spirit’s flight attendants, said: “This is outrageous! We are filing a court objection to Google’s attempt to buy data that has no business being sold.”

The union’s objection focuses in part on employee records rather than passenger files. It has sought restrictions on the sale of flight-attendant data and additional protections if the transaction is approved, arguing that the deal’s privacy terms were written more around consumer and passenger data than around decades of workplace communications, payroll history, and disciplinary files.

After the objection, the bankruptcy court delayed the approval hearing from mid-August to September 9.

What Bankruptcy Does to Ordinary Privacy Expectations

Employees and customers rarely contemplate that, if a company fails, their workplace emails or operational histories could be auctioned as training material for a frontier AI lab. Privacy notices and employment policies typically describe uses by the employer, its vendors, and sometimes successors in a sale of the business. A Chapter 11 liquidation that treats internal systems as a standalone AI-training asset sits at the edge of those expectations.

U.S. bankruptcy law is designed to maximize value for creditors. Digital records that have little worth as a going-concern airline can have substantial worth as raw material for models. That mismatch is why this deal is attracting attention beyond aviation: it is a template. Other failed retailers, carriers, and service companies hold similar archives. Once a court-approved path exists, more estates are likely to test it.

State privacy laws complicate the picture. California and other comprehensive privacy statutes generally treat sale of personal information as a regulated activity and give consumers rights to know, delete, and opt out. De-identified data is often carved out — but only if it meets the statute’s standard and the recipient is contractually barred from re-identification. Whether a large, richly linked workplace archive still qualifies after “reasonable” scrubbing is a factual question, not a slogan. Employee data is treated unevenly across those laws; some statutes cover it more clearly than others.

There is also a consent problem that bankruptcy does not resolve on its own. Workers did not send those emails or sit in those Teams channels in order to train a third party’s AI. Customers who emailed Spirit or appeared in operational workflows similarly did not agree to that secondary use. Exclusion of named passenger files reduces one category of risk. It does not automatically cleanse every message, ticket, or schedule that still describes a person in context.

Practical Implications

For companies that are still operating, the episode is a reminder to treat archives as a future asset — and a future liability. Retention schedules, legal-hold hygiene, and clarity in privacy notices about sale, bankruptcy, and AI training are no longer theoretical. Vendors that process HR, email, and collaboration data should expect more questions about what happens if the customer becomes an estate.

For AI developers, licensed enterprise datasets are an alternative to scraping. They also import the seller’s compliance problems: incomplete de-identification, union and employee objections, and the possibility that a judge conditions or blocks the transfer. Paying for a third-party scrub that the buyer designates and funds, as some reporting on this deal describes, will invite scrutiny of who sets the standard and who audits it.

For regulators and courts, the Spirit sale is an early test of whether de-identification plus a bankruptcy order is enough when the product being sold is the inner life of a company. The September 9 hearing will not settle the broader policy debate. It will decide whether this particular $10 million transfer goes forward, and on what conditions.

Google is not purchasing Spirit’s planes. It is purchasing a record of how the airline worked. That record is valuable precisely because it is granular, longitudinal, and human. Those same qualities are why flight attendants, privacy counsel, and anyone who ever worked inside a large organization should watch what the court allows to leave the estate — and in what form.