If you care about privacy, don’t feed your own personal information into a chatbot, which—among other concerns—could use it to train AI tools. Employees, though, mostly don’t have a choice.
A new, multibillion-dollar industry is dedicated to selling and buying work files and emails to companies seeking fresh sources of data to improve their AI models, and it’s only getting bigger. Data privacy laws tend to focus on consumer protections and rarely consider the rights of workers who create that information. What happens when your doctor’s notes, employee contracts, and Slack gossip are fed into the machine?
An agreement reached this week between Google and a union representing an unlikely workforce—flight attendants—could reveal a more permanent answer by creating a process that might reliably separate the data that workers produce from their own personal information. A judge still needs to approve the agreement next week.
The first-of-its-kind deal is part of the liquidation process for Spirit Airlines, which suddenly ceased operations earlier this year. In August, representatives for the budget airline said in a legal brief that it would be selling decades’ worth of data to Google for $10 million; Google said it would not receive personal information as part of the sale. But the union, the 55,000-member Association for Flight Attendants, argued that Google’s proposed safeguards would not prevent serious privacy violations for Spirit Airlines flight attendants, whose records—including 1.99 billion emails, 664 million Microsoft Teams messages, and employee travel and training information, all dating back to 1986—would be included in the sale.
The revised deal requires Spirit to hire a third-party data de-identification company, called Tonic AI, to exclude information that could potentially be associated with individual consumers or employees. Tonic will also be charged with ensuring that data, once de-identified, cannot be relinked to specific people. Google has promised to maintain the data in this new, de-identified form.
Many sensitive records will be excluded from the sale, including employee medical records, documents related to employee investigations and labor relations, and information about industry-wide safety management systems. In early November, a trade organization representing airlines chimed in on the sale, arguing that including that information could make employees less willing to come forward to report safety issues through voluntary programs, which are widely regarded as critical to the US aviation industry’s strong safety record.
The proposed agreement also includes an opt-out clause. Google can choose to renege on the sale if the data-cleaning process reduces the assets by more than 8 percent. Google didn’t respond to a request for comment.
Association for Flight Attendants president Sara Nelson tells WIRED she’s as surprised as anyone that flight attendants are at the forefront of AI-related workers’ rights issues, which often focus on employees who, willingly or unwillingly, train their machine replacements.
“It’s very hard to replace a flight attendant—you’d need a high-functioning robot, and that seems to be further off,” she says. “What this shows is that this is an issue even in jobs you think are most protected from AI. This is an issue for everyone.”
She said the union will immediately begin to demand that other airlines include worker-data-protection clauses in their labor agreements. “If you’re not willing to meet these standards for your workers, we’re going to shame you,” she says. The Association for Flight Attendants has begun meeting with state-level politicians to craft legislation that would write worker data protections into law.
Ari Ezra Waldman, a professor at the UC Irvine School of Law who studies law and technology, calls the proposed settlement “a pretty good outcome” for Spirit Airlines employees. “From now on, unions will know to negotiate to protect the privacy rights in the event of future sales of data assets,” he wrote to WIRED in an email. He points to another, and maybe easier, alternative: Banning the sale of data altogether. “Privacy protections are necessary now more than ever,” he says.
Tech leaders have argued that workers understand that their information will train AI models, which will ultimately make their lives better and easier. “You have to think of the upside, the abundance that AI can result in,” one executive told WIRED this month.
Nelson says US workers aren’t buying the upside of AI and that they’re part of the wider electorate that seems to be growing warier of the tech by the day. She says the vision of AI abundance is “a fantasyland of people in boardrooms with beanbags who wear hoodies.”
