Spirit Airlines stopped flying earlier this year, and in August Google agreed to pay $10 million for its internal data to train AI. That includes about 1.9 billion emails and 664 million Microsoft Teams messages going back to 1986. The flight attendants' union objected, and the revised deal has a company called Tonic AI strip out anything that identifies a person, with Google paying for that work. A hearing on approval is set for November 17.
A dead company's messages have a buyer
Early AI models learned from public web text. Agents that are meant to do office work need something else: a record of how office work actually goes, who asks whom, what gets attached, how a dispute is settled. Labs now build practice workplaces from real company data, and failed startups are already selling their Slack and email archives for exactly that.
A company in bankruptcy owes money and has to sell whatever it owns. Spirit's price works out to about half a cent per email, before counting the chats and the code. That is nothing to a buyer like Google, but it is real money to creditors, so every failing company with a long mailbox history will be tempted to sell.
The protection came from the one place it could
Shoppers in the Toysmart case could point to a privacy policy. When RadioShack went bankrupt in 2015, regulators applied the same thinking, and the sale was cut back to recently active email addresses, with customers able to opt out. Privacy law has been built around customers for twenty-five years. Workers wrote the messages in this archive, but they were never promised anything about them.
So the union did what a customer-side regulator would have done, and its terms are a usable template. Employee medical records, payroll, drug test results, union grievances and investigation files are left out. Tonic must check that the cleaned data cannot be matched back to people, and Google has promised not to try.
One exclusion shows how far this reaches. An airline trade group argued that selling safety-reporting records would make employees less willing to report problems. The same logic applies in any workplace. People write candidly to HR, to a manager or to a colleague only while they believe the words stay inside the company. If that belief goes, the messages get more careful, and the company loses the honest signal it relied on.
What this looks like in an ordinary company
Imagine a distributor with 60 employees that shuts down. The administrator offers its mailbox archive to an AI lab. Inside are a sick-leave note, a complaint about a supervisor, and a thread about a forklift accident. Removing names does not make these anonymous, because a message about the night-shift lead at one depot points to one person. Without a contract clause, the only safeguard is the buyer's own promise.
The union's president says it will now demand worker-data clauses from other airlines, and it has begun meeting state politicians about writing worker data protections into law. Expect that to spread to other industries. The practical question for any employee, manager or union rep is the one Spirit's staff never got to ask in advance: if this company is sold or closes, what happens to our messages, and who decides?
Company data has never really belonged only to the company. It holds people's working lives, and AI has made that data valuable enough that someone will eventually try to sell it.