Google's $10M for Spirit's inbox and code gives bankrupt firms' data a market price

Price discovery is what actually changed hands. Charge the winning bid across the item counts on the assets schedule and it works out to roughly 1.7 cents per item, or about ten cents an email if the whole price is assigned to the mail spool alone ^{[2] [3]}. Those are cheap enough numbers that volume, not quality, sets the bid.

A wound-down carrier is the seller with the least left to protect. Spirit stopped flying in May as part of its bankruptcy ^[2], so there is no competitor to tip off and no customer relationship left to damage, and the sale sits inside a process whose purpose is converting assets into recovery. The parties with the strongest interest in the contents no longer work there.

On the buy side, Micro1 says it hears from at least 200 interested companies a day and has committed roughly $20 million to enterprise data purchases in the past two weeks, not counting its Spirit offer ^{[12] [13]}. That is about $1.4 million a day ^[4], and at the company's own stated range of $100,000 to $2 million per dataset ^[11], no fewer than ten sellers in a fortnight ^[5]. The material being hunted is the operating residue that had no buyers before: GitHub logs, Slack chats, knowledge bases, support documentation ^[22]. Micro1's founder and CEO Ali Ansari says the purpose is simulated work environments for reinforcement learning, with agents installed as employees at fictitious companies generated from anonymized real business data ^[14].

The safeguards on offer are process, not property. Court records show that under the Google deal any personally identifiable information goes through a third-party deidentification agent, and customer lists are excluded ^[8]. Ansari says an automated pass strips email addresses and ID numbers before models see the data, and raw files are usually deleted within 30 days ^[15]. Neither answers linkage, since anonymized records can still be tied to individuals through context or speech patterns ^[18]. The buyer's own perimeter is part of the trade too: Mercor, which also bid, faces litigation after disclosing a March security incident involving sensitive information about the experts it pays to train AI ^[17].

Alexandra Mateescu, a researcher at Data and Society's Labor Futures initiative, makes the point that survives all of this: workplace documents belong to employers rather than the staff who wrote them, and the agents being trained on them are aimed at the jobs of those same people, without consent or payment ^[19]. Employees have few legal levers once an estate is selling ^[20]. The one that exists is contractual, since some union agreements already contain provisions covering employee data rights ^[21], and that has to be bargained while the airline is still flying.