Google pays $10M for a dead airline's training data

Share
Google pays $10M for a dead airline's training data

When the open web runs dry, bankruptcy court starts to look like a data vendor.

Google won a $10 million auction for the deidentified business data, software code, and operations records of collapsed low-cost carrier Spirit Airlines, and it says the haul will help improve its products and AI models. An August 14 notice in the US Bankruptcy Court for the Southern District of New York describes a private corpus no crawler could have built: 100 million emails, 500 million Microsoft Teams chats and collaboration records, plus files on revenue, aircraft operations, employee productivity, audits and fraud, marketing campaigns, human resources, strategy, and project management. Court records also list pricing from 7.2 billion competitor flights, an estimated 7.5 billion passenger transaction records going back to 2008, about 30 million lines of code with development metadata and software models, and more than 175,000 employee records dating to 1986. Google beat a $7.5 million backup bid from AI recruiting firm Mercor.

This is not a travel product. It is a lab buying two decades of how a real company actually ran — the messy, proprietary residue that public web scrapes never captured. Florida-based Spirit Aviation Holdings shut down on May 2 in its second Chapter 11 in two years, carrying about $8.1 billion in debt and laying off roughly 17,000 people. A month later its LaGuardia takeoff and landing slots sold to JetBlue for $58.5 million. The emails and code went cheaper, which tells you what the market currently thinks an airline's operational memory is worth next to a gate at LGA.

Google is careful about the privacy line. The purchase excludes Spirit's 97.5 million passenger profiles and an estimated 50.2 million Free Spirit loyalty records, and the company said the data will be "rigorously scrubbed of any personally identifiable information by a third party before receipt." In a statement Monday it put the deal in the blandest possible box: "We acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models." Scrubbed is not the same as harmless. Five hundred million workplace chats and 175,000 employee files, even stripped of names, are still a map of how people priced flights, handled fraud, argued in Teams, and wrote the software that kept a carrier in the air.

The deeper tell is who else showed up. Mercor is not an airline. It is an AI talent shop that just tried to outbid Google for the same inbox. That is a market forming: when a company dies, its internal communications become a training set, and the bidders are model builders, not competitors in the dead firm's industry. We covered the other end of this hunger this afternoon — The AirTag trail ends at Amazon's book-scanning warehouse — where used books are being cut apart for text. Spirit is the enterprise version of the same bet. The web was the first corpus. Books were the second. Bankrupt companies may be the third.

A sale hearing is set for August 19 before Judge Sean Lane. If the sale clears, expect the next Chapter 11 to list "deidentified enterprise dataset" next to the gates and the spare engines.

What to watch: whether Judge Lane signs off, and whether Mercor or another model shop bids on the next failed carrier.

Should a dead company's internal emails be sold to train someone else's model? Tell us in the comments.

Read more

Byteification turns Qwen3 and Llama 3 into byte-level models

Byteification turns Qwen3 and Llama 3 into byte-level models

The tokenizer is the last piece of the language-model pipeline that was never learned end to end — and a Nature paper out this week shows it can be swapped out for less than one percent of a pretraining budget. Researchers have retrofitted four open subword language models into true byte-level models using less than 1% of a typical pretraining budget, and the result — published 7 October in Nature — closes the performance gap that has kept byte-level models a niche curiosity for nearly a decad

OpenAI's dots go mobile and can start your Codex work

OpenAI's dots go mobile and can start your Codex work

One product update and two benchmark papers this hour, all circling the same question: how much of the work can you actually hand over, and what happens to the part you didn't? OpenAI's dots — the always-on agents that get their own cloud computer — can now be created and configured entirely from the ChatGPT app on iOS and Android, and a dot can start Codex work or pick up an existing Codex thread on your behalf. The October 9 release, billed as the first of a fresh run of updates for dots, mo

Coding agents as a hospital: Cockroach's five-month results

Coding agents as a hospital: Cockroach's five-month results

The agentic-coding debate keeps being framed as speed versus trust. Cockroach Labs just published the most detailed trust-first ledger yet — five months of numbers, including the parts that didn't work. Cockroach Labs ran its codebase like a teaching hospital for five months and says the pipeline merged well over a million lines of code with seven total reverts. In the setup the company calls MOLT Sinai, GitHub issues are patients, merging is discharge, and the humans in charge are Chiefs of M

Today in AI — October 10, 2026

Today in AI — October 10, 2026

A day where the boring machinery took center stage: proof checkers, order books, warehouses and the humble text message — plus fresh arguments about who gets to declare any of it safe. Models & Research * Mathematicians get a reliability primer for the Lean Theorem Prover. A guest post by Thomas Hales on Terence Tao's blog walks through what Lean actually guarantees — and devotes itself to the "Summer of Soundness Bugs," the string of kernel bugs found this summer that let false proofs thr