Court stalls Google's $10M Spirit data deal over worker privacy

Why it matters

Spirit's estate is testing whether a bankrupt employer can turn years of employee communications into AI inventory after removing identifiers. The court's terms could shape future data sales.

Video version

A federal bankruptcy court delayed approval of Google's $10 million purchase of Spirit Airlines' digital records after former flight attendants challenged the privacy terms, according to The Wall Street Journal. The hearing, originally scheduled for August 19th, was moved to September 9th after the Association of Flight Attendants-CWA filed its objection, Reuters reported.

The dispute reaches beyond names, email addresses and other obvious identifiers. Spirit's proposed sale covers decades of workplace history, including payroll and training files, crew schedules, internal communications, software repositories and employee productivity records. Google says the material could improve its products and AI models.

The union's argument is that de-identification answers the wrong question. Removing information that directly identifies an employee does not remove confidential details about discipline, medical accommodations, performance, union activity or internal grievances. Those details can remain sensitive even when attached to a pseudonym or employee number.

The dataset is mostly workplace history

Spirit selected Google as the winning bidder at an August 14th auction, with Mercor.io named the backup bidder at $7.5 million, according to the auction notice and proposed sale agreement.

The asset schedule shows why the flight attendants intervened. Spirit proposes to transfer about 100 million emails across 80,000 accounts, 500 million Microsoft Teams items, 17.1 million OneDrive items and 20.6 million SharePoint items. The package also includes 175,658 employee records dating to August 1986, 3.4 million payroll records, 148,018 employee tax forms, crew training data and more than 5 million crew pairings.

Google would also receive 516 source-code repositories containing approximately 30 million lines of code, commit histories, bug reports and development discussions. Operational and commercial records cover aircraft logistics, pricing, revenue management, audits, fraud, project management and customer-service workflows.

Spirit excluded 97.5 million passenger profiles and 50.2 million Free Spirit loyalty records from this transaction. The dividing line leaves customer databases outside the deal while admitting much of Spirit's employment record.

Google told Axios that a third party would rigorously scrub personally identifiable information before Google received the dataset. The proposed agreement also says Google must keep the information de-identified and refrain from intentionally associating it with a person or household.

De-identification leaves context intact

The AFA-CWA objection focuses on what those protections preserve. The agreement requires the de-identification process to maintain "referential integrity" across the dataset, allowing records from different systems to remain linked.

That linkage has technical value. An AI system can study how an operational event moved through scheduling, communications, payroll and management systems. It also creates the risk identified by the union: a small employee group may remain recognizable from its crew base, assignments, training history, grievances and conversations, even after names are removed.

The agreement only prohibits Google from intentionally reconnecting the information to individuals. It allows Google to transfer the de-identified data to third parties that accept the same contractual restriction. Flight attendants would have no third-party beneficiary rights under the agreement, according to the union's filing.

AFA-CWA wants Spirit to remove all flight attendant information. Its fallback request would require a separate review for personnel, payroll, medical, leave, disciplinary, performance and union records. The union also wants restrictions against using the data to profile, score or draw conclusions about individual flight attendants or identifiable groups.

The data has not been delivered to Google, and the de-identification protocol has yet to be designed, AFA-CWA said in an August 18th update.

The auction priced privacy into the bid

Spirit's bidding record shows that privacy controls affected the auction's economics. Google opened at $5 million and agreed to pay for a third-party de-identification process. Mercor later offered $10 million if it could perform that work with its own tools, according to a declaration filed by Spirit's investment banker.

Spirit's advisers chose Google's $10 million proposal and designated Mercor's $7.5 million version, which used Google's third-party process, as the alternate bid. That decision placed a $2.5 million difference between the preferred backup structure and Mercor's higher, self-scrubbed proposal.

The data itself offers Google something public web crawls cannot: a connected record of how a large organization scheduled workers, priced inventory, handled disruptions, wrote software and communicated internally. Spirit's collapse does not diminish that training value. It made the records available as an asset.

Google has been shortening the cadence between model releases as it competes across coding and workplace automation. RuntimeWire reported on August 13th that Google replaced Gemini 3.6 Flash after three weeks while cutting prices through the end of 2026. Spirit's records would give Google a proprietary enterprise corpus for further model and product development, although Google has not specified which systems would use it.

A bankruptcy test for AI training data

The legal structure helps explain the gap the union is challenging. Section 363 of the U.S. Bankruptcy Code contains specific restrictions for selling personally identifiable information when a debtor promised customers it would not transfer that information. The code's definition of personally identifiable information is tied largely to information supplied by an individual while obtaining a product or service.

Employee communications occupy a less explicit position in that framework. AFA-CWA is asking the court to use its authority over the sale terms to impose protections aimed at workplace confidentiality rather than relying on a consumer privacy standard.

The September 9th hearing will determine whether Google can proceed under the current structure or must accept a narrower dataset and additional use restrictions. A ruling that approves the sale without an employee-specific screen would give other bankruptcy estates a path to package workplace systems as AI assets. Requiring a carve-out would establish that removing names alone is insufficient when the value of a dataset comes from preserving years of linked human behavior.