US data protection

Spirit's Data Sale Shows Bankruptcy Law Protects Customer Records but Leaves Employee Archives to Negotiation

A judge approved Google's $10M purchase of Spirit Airlines' employee data for AI training. The safeguards came from a union deal, not from any statute.

Spirit Airlines Data Sale to Google People of Internet Research · US $10M Sale price Google's winning bid, approved by … ~100M Emails in dataset Estimated volume, plus roughly hal… 121 Lawmakers objecting Members of Congress signed the Oct… peopleofinternet.com
Spirit Airlines Data Sale to Google People of Internet Research · US $10M Sale price ~100M Emails in dataset 121 Lawmakers objecting peopleofinternet.com

Key Takeaways

On 17 November 2026, US Bankruptcy Judge Sean Lane approved the sale of defunct Spirit Airlines' internal data to Google for $10 million. Google has said it intends to use the data for AI model training. According to The Record, the dataset includes an estimated 100 million emails and half a billion Microsoft Teams messages. Judge Lane found that the "procedures for the deidentification, transformation and preparation of the data" comply with applicable data protection laws.

The outcome is better than it could have been. The route to it shows a structural gap in US privacy law, and that gap deserves attention without panic.

The strongest case for blocking the sale

Critics have a serious argument. Employees wrote those emails and messages for colleagues and managers, not for a model-training corpus. In an October letter to the CEOs of Google and Spirit, 121 federal lawmakers, led by Rep. Steven Horsford and Sen. Elizabeth Warren, wrote that "conventional de-identification safeguards may not be sufficient" and that removing names and email addresses "does not necessarily make a dataset anonymous" (The Record). Workplace messages are rich in context, such as project names, writing style, and who talks to whom. Those details can re-identify a person even after direct identifiers are stripped. The workers had no say in the sale, and the airline's collapse was not their doing.

The contract gives that worry some support. The Next Web's reading of the 14 August sale agreement found that the buyer designates and pays the scrubbing agent. It also found that the certification must hold "while preserving referential integrity across the data set". That means records belonging to one pseudonymous person can still be linked to each other. That feature is useful for AI training and awkward for privacy.

Where the legal protection actually sits

US bankruptcy law does have a privacy mechanism, but it is aimed at consumers. Section 363(b)(1) restricts the sale of personally identifiable information when the debtor's privacy policy promised not to share it. Absent compliance with that policy, a court must first hear from a consumer privacy ombudsman. Federal Rule of Bankruptcy Procedure 6004 requires the US trustee to give notice of that appointment at least seven days before the hearing. The FTC has used the mechanism for years. In 2015 it asked a court to require that RadioShack's customer data not be sold as a standalone asset. The data was to go only to a buyer in substantially the same line of business, bound by the original privacy promises.

That framework rests on promises made to customers. Employee records involve no public privacy policy to enforce and no FTC deception theory to invoke. Spirit's customer data was reportedly excluded from this sale, so the consumer-protection machinery had little to grab. Employee protections had to come from somewhere else.

What the union won

They came from litigation and negotiation. On 9 November the flight attendants' union withdrew its objection. According to the Association of Flight Attendants' release, the protections include:

These are meaningful terms. They are also contractual, and they exist because an organized, well-resourced group was at the table. A dispersed workforce with no union would have had a trustee and a buyer negotiating over its messages. The protections tracked bargaining power, not any legal entitlement.

Why a ban on the sale would be the wrong lesson

A blanket prohibition on selling workplace data in bankruptcy would be a blunt instrument. Estates exist to maximize recovery for creditors, and those creditors include many of the same former employees owed wages and benefits. The bidding was real. According to The Next Web, Google opened at $5 million and closed at $10 million, with Mercor as a backup bidder at $7.5 million. A rule that zeroed out the value of a data asset would not make the data vanish. It would push it to quieter, less scrutinized disposals, or leave it in unmanaged estates where breach risk grows.

The better response is proportionate and procedural. Congress or the courts can build on what this case demonstrated works:

None of this requires a new federal privacy regime or a ban on training AI with lawfully obtained data. It requires closing a narrow gap in a process that already exists.

The open question

The deal is, per The Record, the first of this scale in the AI era, so the precedent matters. Bhatnagar of the privacy startup Cloaked warned that a promise not to reidentify "is supposed to follow the data," yet datasets get copied, licensed, or breached. That is the fair residual concern. Contract terms bind Google. They do not bind everyone downstream. If an independent assessor can verify that reidentification risk is low, and the court can enforce the terms, the sale is defensible. If not, the next estate will not have a union as organized as this one. Policymakers should write the safeguards in before then, so workers do not depend on having a strong negotiating position.

Sources & Citations

  1. The Record: Judge signs off on Spirit data sale
  2. The Record: Lawmakers warn of Google-Spirit AI training deal
  3. The Next Web: Google picked and paid for the anonymising firm
  4. AFA-CWA release on data privacy protections
  5. FTC: RadioShack consumer data request (2015)
  6. Fed. R. Bankr. P. 6004 (Cornell LII)