Google’s Spirit Data Deal Raises $10MN Question: Should AI Developers Be Allowed to Buy Enterprise Data for Training?

Summary

Google has emerged as the winning bidder in a bankruptcy auction for a large collection of de-identified operational data from Spirit Airlines, offering $10 million to acquire records that include employee communications, workflow logs, HR files and productivity metrics. The dataset encompasses more than 100 million emails, 500 million Microsoft Teams items and millions of payroll and time-card entries. A competing AI training firm, Mercor, placed the second-highest bid at $7.5 million. The transaction still requires approval from the U.S. Bankruptcy Court for the Southern District of New York, with a hearing now scheduled for 9 September after an objection from the Association of Flight Attendants-CWA.

The proposed sale agreement calls for a third-party agent to process the data so that it cannot be linked to individual consumers and to exclude categories of personal information protected by data-protection statutes. The process must nevertheless preserve referential integrity across tables and documents. The flight attendants’ union contends that these safeguards were designed primarily for customer records, whereas the bulk of the material concerns employees. It argues that removing names or identifiers does not eliminate the sensitivity of internal performance data, collaboration logs and operational communications, and that employees never consented to the use of their work product for third-party AI model training.

The case illustrates how enterprise data generated during routine operations is gaining value as training material for AI systems. Bidders initially sought customer lists but later shifted to operational datasets once personally identifiable information was excluded. The union has asked the court to bar any transfer, licensing or downstream use of flight-attendant records by Google or subsequent recipients. The proceedings may clarify whether bankruptcy proceedings allow operational data to be treated as a conventional corporate asset and what standards of anonymization and consent should apply when employee-generated content is repurposed for model development.

Why it matters

This article explores the legal and ethical boundaries of repurposing enterprise and employee data for AI training. It touches on core GDPR concepts like anonymization, pseudonymization, and legitimate interest, offering crucial insights for Dutch organizations establishing AI data governance frameworks.