Micro1 Challenges Google With $12.5 Million Bid for Spirit Aviation’s AI Training Data

AI training-data company Micro1 has made a last-minute bid to acquire data assets from bankrupt Spirit Aviation Holdings, offering $12.5 million in cash—25 percent more than Google’s competing proposal.

In a competitive bidding filing, Micro1 said it would resolve objections related to the sale and avoid creating additional delays in Spirit’s bankruptcy proceedings. The company also indicated that it would anonymize the data as part of the acquisition.

Spirit’s debtors, however, previously rejected Micro1’s approach during the auction. They selected a lower bid because the winning proposal required the data to be anonymized by an independent third party. That safeguard was viewed as more reliable, particularly because the dataset may contain sensitive proprietary information.

Micro1’s promise to resolve all outstanding concerns may also be narrower than it initially appears. For Springshot, one company that has challenged the sale, Micro1 has not committed to separating intellectual property, trade secrets, or other proprietary information from the dataset.

Instead, Micro1 has said it would not use Springshot’s data to develop competing products or sell the information to rival companies. That commitment may not fully address concerns about how third-party data could be handled, anonymized, or incorporated into AI training systems.

Micro1’s lawyers did not respond to Ars Technica’s request for comment. The company’s Data Lab team is expected to attend a future bankruptcy hearing and argue that Spirit’s debtors should accept its higher offer.

Kreutzkamp, a representative of Springshot, said he appreciated Micro1’s acknowledgment of the company’s concerns but maintained that Springshot’s primary objective is to remove its data from Spirit’s dataset entirely.

Lawyers representing Springshot have contacted Spirit’s legal team and plan to continue defending the company’s sensitive information in upcoming hearings. They believe the bankruptcy court will seriously consider their objections, although the court has not guaranteed that the data will be removed before approving the proposed sale.

The dispute highlights a growing legal challenge involving AI training data: how can information belonging to multiple companies be identified and removed from a dataset before it is transferred to a new owner?

“I think it’s a very complex problem when it comes to extracting data that is owned by other people,” Kreutzkamp said. He argued that the issue is especially urgent because “once you throw data into an AI model, it’s impossible to track what the model did with it.”