Development
Panic builds over bankrupt Spirit’s looming data sale to Google
September 10, 2026 Development Source: Ars Technica
Share this article
“Bankruptcy cannot become the new land grab for AI,” Kreuzkamp told Ars. “The possession of IP is not ownership.”
Asked for comment, Google’s spokesperson declined to discuss mounting objections and repeated a prior statement provided to Ars that did not address any of the concerns raised.
“We acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models,” Google’s spokesperson said. “We will not receive any personal information from this dataset.”
It seems likely that a wide range of Spirit vendors could be in the same spot as Springshot, Kreuzkamp suspects, and his firm is not the only one complaining.
Citing Springshot’s concerns in a separate objection, the International Aero Engines LLC and the IAE International Aero Engines AG alleged that the dataset might also include its proprietary commercial information, technical data, and financial data. Both vendors cited confidentiality provisions in Spirit agreements that appear to have been completely ignored in the push to get the data sale approved.
Top concerns are that proprietary data could be transferred to Google without consent, giving firms no chance to guard against Google—or a third party later gaining access to the data—infringing ownership rights and exposing trade secrets. Both vendors said they risked irreparable harm through the sale to Google and then again through possible resales to future third parties.
For Springshot, it also seems possible that Google could use its valuable IP to create a rival product. In August, just a week before the controversial Spirit auction, Europe’s largest airline, Ryanair, announced a five-year partnership with Google. Under the deal, Ryanair will share operational data to improve Gemini Enterprise tools. That “is precisely the function Springshot served for Spirit,” the startup said in a footnote in its objection, while alleging that “Springshot’s data may be among the most AI-relevant assets to be purchased.” As Springshot argued:
“Should the sale proceed without such safeguards, and should Google obtain Springshot’s intellectual property and ingest that data, Springshot would forfeit the value of its technology it has spent 15 years developing. That existential threat is only amplified by the fact that Google has recently announced its desire to operationalize its own AI platform for airline operations, placing it in direct competition with Springshot.”
Adam Schwartz, a privacy litigation director for a digital rights nonprofit called the Electronic Frontier Foundation, told Ars that allowing Google to buy this dataset via bankruptcy proceedings is unusual.
“This is the first time I am aware of so public a bankruptcy proceeding regarding whether a bankrupt company may sell off the personal data it has amassed as an asset in bankruptcy,” Schwartz said. “It is also the first time I am aware of a company seeking to sell its employees’ data (as opposed to its customers’ data) as an asset in bankruptcy.”
Kreuzkamp is worried that because bankruptcy courts have not deeply considered this issue before, the system is just not set up to protect vendors like his company. In the past, it was straightforward to determine who owned an asset. Consider how Spirit is selling airplanes, which bills of sale clearly demonstrate that it owns, he said. His experience suggests that when it comes to digital assets, courts don’t have the proper notice provisions that are necessary to ensure vendors can advocate for IP to be segregated from sales, should purchases like Google’s become the norm.
“Our main concern with everything is if you want startups to continue building, you need to protect what they built,” Kreuzkamp told Ars. “If they spent 15 years building IP, that IP should not be misappropriated because their customer went bankrupt.”
Troublingly, the sale agreement only required de-identification of customer personal data, and former Spirit workers worried that Google’s AI system may combine confidential worker data in ways that may expose information that they might not want public. As groups representing machinists, aerospace workers, and transport workers noted, workers aren’t even aware of all the risks, since “the proposed sale raises novel technological and legal concerns on very short notice.” Seemingly, not even Spirit knows the “ultimate uses” to which the data “may be put after it is sold and processed through artificial intelligence,” they warned.
Schwartz told Ars that Spirit should have gotten consent from workers before selling 80,000 email accounts, 100 million emails, 20 million SharePoint documents, and 500 million Teams messages.
“My top concern is that data privacy is a fundamental human right, enjoyed for example by employees in relation to their employers,” Schwartz said. “This includes the norm that data collected from a person for one reason should not be used or disclosed for another reason without first getting their opt-in consent. Therefore, it is the Spirit Airline employees themselves who should be making the decision whether they want to bear the risk to their privacy of their emails being turned into AI training data.”
Some former workers are alarmed that the data sale could possibly harm the public.
Most ominously, the Air Line Pilots Association, International warned that if Google’s AI system re-identifies pilots in the data set, airways could become more dangerous.
Flight safety programs depend on pilots feeling comfortable sharing in-depth incident reports without fear that confidential information may become public, the Federal Aviation Administration has found. Therefore, any “disclosure of confidential flight operations data through the Sale Agreement can be expected to chill voluntary compliance by pilots across the industry as it becomes known that such data can be sold to third parties with no expertise or interest in aviation safety and presumably included in databases accessible by the public using artificial intelligence software,” the association warned.
Last week, an AI training data lab, Micro1, put in a last-minute competing bid that it claims resolves all objections to the data sale. Outbidding Google by 25 percent, Micro1 offered $12.5 million in cash, while promising that if debtors pick Micro1 over Google, the bankruptcy proceedings can avoid any further delays.
However, Micro1’s plan notably includes de-identifying the data in-house, which Spirit debtors were opposed to in the auction process. Spirit even chose a lower second bid should the Google sale fall through because the de-identification was completed by independent third parties. It seems, then, that even if the late bid is considered, it’s less likely to tempt debtors. Additionally, Micro1’s claims to resolve all objections may be somewhat exaggerated. In Springshot’s case, Micro1 does not promise to segregate IP, trade secrets, or proprietary data. Instead, Micro1 only says that it will not use Springshot data to create a rival product or ever sell the data to a would-be rival.
Ars could not reach Micro1’s lawyers for comment, but the data lab said it plans to attend the upcoming hearing to entice debtors to take their offer.
Kreuzkamp told Ars that he appreciates that Micro1 acknowledges Springshot’s concerns, but his objective remains scrubbing Spirit’s dataset of any data that Springshot owns.
His company’s lawyers have been in contact with Spirit debtors’ legal team and plan to defend Springshot’s proprietary data at the upcoming hearing. They feel confident the court will take their concerns seriously. But there are no guarantees, as the bankruptcy court prepares to approve this unusual sale. Kreuzkamp said it remains unclear what process a bankrupt company like Spirit should be taking to “ensure that it extricates everything that is proprietary from a dataset.”
“I just think it’s a very complex issue when it comes to extricating the data that’s owned by somebody else,” Kreuzkamp said. However, now’s the time to do it, he suggested, since “once you throw data into an AI model, it’s impossible to trace what the model did with that.”