Google Bought a Dead Airline's Inbox for 10 Million Dollars
Bankruptcy filings show Google won an auction for Spirit Airlines' internal data: around 100 million emails, 500 million Teams messages, code and pricing records. Customer and loyalty data is excluded.
Spirit Airlines stopped flying in early May after its second bankruptcy in under a year. What was left included the ordinary digital residue of a company: a mail server, a chat archive, a code repository, decades of pricing spreadsheets. According to filings in the bankruptcy proceedings, Google bid 10 million dollars for that pile and won it at auction. The AI data company Mercor bid 7.5 million and lost.
The package is large and unglamorous. Reports put it at roughly 100 million emails and around 500 million messages from Microsoft Teams, together with software, source code, pricing models, booking curves, flight behaviour data, inflight purchase and Wi-Fi sales records, refund histories and travel package information. Employee records reportedly go back to 1986. Google says all personally identifiable information will be stripped out by a third party before the data is handed over. The deal explicitly excludes customer and loyalty data, which covers 97.5 million passengers, 52.4 million loyalty members and 740,000 co-branded cardholders. Google says the corpus will help improve its products and models.
Why would anyone want a defunct airline’s group chat? Because of what it contains that the open internet does not. Almost every large language model has been trained on the public web, which is full of finished, polished, published text. What it is thin on is the messy middle of how work actually gets done: the thread where four people argue about a pricing rule, the half-written escalation, the moment somebody explains a policy to a new colleague in plain words. If you are trying to build models that handle real business workflows rather than write blog posts, that material is the scarce ingredient. Ten million dollars for a decade of it is, by the standards of this industry, cheap.
There is a fair caveat and it is about the word anonymised. Stripping names and account numbers from a corporate mailbox is genuinely hard. Emails are full of context that identifies people without naming them: a project, a date, a role, a distinctive turn of phrase. Independent researchers get no access here, so the quality of that scrubbing is something we take on trust. The broader precedent is the part to watch. Corporate data is now an asset that gets auctioned in bankruptcy court, which means the people who wrote those emails have no say in where they end up.
What this means for you: practically, nothing changes today, and if you flew Spirit your booking history is not in the sale. The useful thought is about your own work. Anything you write in a company system belongs to the company, and companies can be sold or wound up. That was always true, but the resale value of internal chatter used to be roughly zero. It is not zero any more, which is a reasonable thing to keep in mind the next time you type something into a work chat that you would not want a model trained on.
Sources
Source: https://skift.com/2026/08/17/google-scoops-up-spirits-data-in-bankruptcy-sale-to-train-ai/
Reddit Is Testing a Play Button That Reads Threads Aloud in AI Voices
A limited experiment turns selected text posts and their top comments into short narrated videos, labelled 'Real conversation voiced by AI'. The original thread stays exactly where it was.
