Technology
Competing bid emerges as Spirit's data sale to Google faces last-minute legal challenge
An AI training data firm offers 25 percent more than Google to acquire bankrupt Spirit Aviation's datasets, but raises new concerns about protecting third-party intellectual property.
Ars Technica
A data-focused AI company called Micro1 has submitted a competing offer to acquire Spirit Aviation's datasets ahead of a bankruptcy auction, significantly outbidding Google's proposal with a $12.5 million cash offer. The move comes as legal objections intensify over how proprietary information from third parties would be handled in the sale.
The Competing Offer#
Micro1's bid exceeds Google's by 25 percent and includes assurances that the transaction could proceed without additional delays if the bankruptcy trustees select Micro1 instead. The firm claims its proposal addresses all outstanding concerns raised during the auction process—though that assertion appears to face substantial pushback.
The De-identification Problem#
One core dispute centers on how sensitive data gets cleaned before transfer. Spirit's debtors had explicitly rejected de-identification done in-house, instead selecting Google's higher-ranked bid partly because an independent third party would handle that work. Micro1's plan to de-identify the data itself undercuts what trustees saw as a key safeguard against misuse. This approach mirrors concerns that initially made Spirit resistant to other options.
Protecting Third-Party Data#
A particular sticking point involves Springshot, a company whose proprietary material may be embedded in Spirit's datasets. Micro1 has offered limited protections—pledging not to use Springshot's information to build competing products or to sell it to rivals—but stops short of segregating trade secrets and IP separately, as Springshot's leadership demands.
Springshot executives remain unconvinced. Their legal team plans to present arguments at the upcoming hearing, arguing that complete extraction of third-party proprietary content from the dataset is the only acceptable outcome. As one executive noted, once data enters an AI model's training process, determining exactly how it influenced the final system becomes impossible to track.
