Google agrees to scrub Spirit Airlines archive before $10 million purchase

Work emails, Teams messages and employee documents can outlive a company—and become a commodity for AI training. According to WIRED, Google agreed to buy the archive of the shuttered Spirit Airlines for $10 million. The flight attendants' union intervened in the deal and secured conditions: sensitive documents will be excluded, and the remaining material must undergo independent anonymization. A court still needs to approve the agreement.

For developers and founders, this is a familiar problem: a large corporate archive looks like a useful dataset until it turns out to contain medical information, workplace disputes and correspondence from people who never chose to take part in model training. Deleting a column of names is not enough for this task.

What was slated for sale

The deal is part of Spirit Airlines' liquidation. The collection described by WIRED includes 1.99 billion emails, 664 million Microsoft Teams messages, and information about employee travel and training. The archive contains material dating back to 1986; that does not mean every type of data was collected throughout that period.

Google had said personal information would not be included in the purchase. However, the Association of Flight Attendants, which represents 55,000 flight attendants, considered the original safeguards insufficient to protect Spirit workers. That figure refers to the union's entire membership, not the airline employees whose information appears in the archive.

Under the revised terms, Spirit must hire Tonic AI. The contractor must exclude information that can be linked to specific consumers or employees and verify that they cannot be reidentified after anonymization. Google commits to keeping the data in that form.

Employee medical records, internal investigation documents and labor relations documents will be excluded from the sale. The same applies to information about aviation safety management systems. The risk here extends beyond privacy: the prospect of safety reports being sold could discourage workers from submitting them.

Why removing names is not enough

Imagine an email: "After Friday's flight, I'll miss training because of a procedure at the clinic." Even without a signature, the combination of the route, date, training schedule and other messages could reveal who wrote it. This is a hypothetical example, not a quotation from Spirit's archive.

That is why the requirement to prevent reidentification matters more than a promise simply to erase obvious identifiers. Corporate data often derives its usefulness from context: who is writing to whom, what situation they are discussing and what decisions they are making. But that same context can reveal someone's identity.

For a team buying data or preparing its own correspondence for AI, the practical question is: what connections remain after the cleanup? Reviewing individual documents does not fully answer it. Teams need to account for the possibility that information from several emails can combine into an identifiable story.

The article does not disclose Tonic AI's methodology, verification results or residual risk. For now, therefore, the agreement sets out processing requirements rather than demonstrating the effectiveness of a particular anonymization technology.

Privacy now has a price tag

The terms contain a significant caveat: Google can walk away from the purchase if the cleanup reduces the volume of the assets by more than 8%. This gives the buyer the right to exit the deal. The text provided does not imply that every employee has been given an individual option to opt out of having their data transferred.

The threshold illustrates the commercial tension surrounding such archives. The buyer needs sufficiently rich material, while protecting people may require deleting a substantial portion of it. Assuming in advance that the entire purchased collection will be suitable for training is a poor basis for estimating a project's cost.

The practical takeaway for product managers and founders: before buying an archive, define the categories of documents to be excluded, the anonymization requirements and the consequences of losing useful data. Those terms affect both the budget and whether the purchase can be used at all.

The union intends to seek similar provisions in agreements with other airlines. For now, this is a single agreement awaiting court approval. But it already poses a concrete question for the data market: how much of a valuable workplace archive will remain once information about people has actually been removed?