If Google will pay $10 million for old emails and business records, what is your personal data worth in the AI era to come?

If you ever doubted data is the new oil, check out the $10 million that Google has put on the table to snap up every email that Spirit Airlines employees ever sent - and then consider the precedent that this deal is likely to set for future bankruptcies in the US and your privacy! For context, back in May, US budget airline Spirit Airlines hit the tarmac for the last time, filing for bankruptcy after months of speculation surrounding its fiscal viability. As is customary in such scenarios, the receivers handling the bankruptcy proceedings set about a corporate fire sale to try to pay down some of the firm’s debts. What’s different here in the AI age is that Big Tech turned up at the auction and bid for data. Not physical assets, not property, data - information held in the airline’s systems that could be used by the company for frontier model training. And that’s a lot of data - roughly 100 million emails, 500 million Microsoft Teams messages, 17 million OneDrive files, and 20.6 million Sharepoint items. Then there’s all the employee records dating back to the mid-1980s, 34 years of internal business records and correspondence, and 516 code repositories build by Spirit’s internal IT teams. Data from passenger profiles or information captured from the Free Spirit loyalty program are explicitly excluded from this deal and Google has promised that any material it buys will be cleansed by an independent third party to remove any personal identifiers. But nonetheless, what Google gets for its $10 million is essentially an enterprise digital model of how an airline operates - albeit one that ultimately didn’t operate well enough to survive, but still… - and the long-term value of that is almost certainly worth a lot more than the $10 million it’s ready to pay. Specialist data In the AI age, specialized data to build new frontier models and train up AI systems is going to become every more valuable and ever more a competitive differentiator in an already over-crowded and febrile market. Oracle’s Larry Ellison summed up the realities of the situation a few months ago when he said: > All the Large Language Models—OpenAI, Anthropic, Meta, Google, xAI—they’re all trained on the same data. It’s all public data from the internet. So they’re all basically the same. And that’s why they’re becoming commoditized so quickly...The future lies in leveraging private enterprise data. Now, critics will suggest that all too many AI model firms have approached the need for data to build out their model capabilities by copyright infringement and wholesale theft and there’s plenty of evidence - and pending litigation - to support such a thesis. So on the one hand, perhaps it’s important to note that Google, on the face of it, has gone about this in a more business-like manner, being ready to pay to buy in other people’s knowledge. On the other hand, the knowledge that it’s buying was provided by people over a period of some decades when they could not possibly have suspected that it would be used in since a manner. By 2036, if someone’s trying to buy the previous five years worth of data from a firm then there’s possibly a case to be made that everyone ought to be fully aware of what might happen to any data handed over to an organization, unless opt-outs are opted-into or there’s some significant movement on the regulatory tectonic plates. But right now, this is data that was gathered before gen AI was a thing. Precedent-setting The players involved here are high-profile enough that this is getting a lot of attention in its own right. But the really important thing is not what happens in this individual case, but rather the precedent that it sets. This will not be the last time this sort of data fire sale occurs. The value of corporate data is not going to fall, any more than the voracious appetite from model firms for access to that data is going to decline notably either. So this or something like it will happen again - and probably with increasing frequency. Google had to fight off competition from a dedicated AI training firm to secure its deal, and while the matter rumbles though the courts, another bidder has emerged to try to gazzump the existing Google offer with a $12.5 bid for its own deal. The demand is clearly there. And while this Google deal explicitly excludes customer-specific data, that can’t be guaranteed to be the case in future examples of such actions. Imagine in five years time, it happens to an organization you’ve worked for or engaged with. Your emails, your Slack messages, your chatter on the corporate intranet, all gobbled up to become so much AI training fodder. As it, the Association of Flight Attendants (AFA) has raised objections to the proposed purchase, with much of the data involving their members. The Association argues in a legal filing that the terms of the sale do not appears to have safeguarding provisions in place to protect confidential workplace data, such as payroll records and internal communications, in the same way that customer-specific information will be: > The privacy architecture of this transaction is consumer-facing; its payload is disproportionately employee-facing. Hence, the employee data is far more confidential than the customer data, yet receives far less protection. The union wants the deal ruled out unless flight attendant data is removed from any data transfer.Another option would be the setting up of an independent review process applied to labour and disciplinary records before anything changes hands, along with clearer details on how material might or might not be passed onto third parties at a later date beyond the initial transfer. The sale has been put on hold by the bankruptcy court until 9 September to give time to rule on AFA’s argument and plea. Consequences It will be interesting to see how this plays out in the US where, as diginomica has noted so many times over the years, legislative and political attitudes towards data privacy and protection are far laxer than, for example, in Europe. Could this sort of scenario occur if Spirit was a German airline? GDPR would surely create enormously high barriers to overcome before any such deal could take place - the data at the heart of the Spirit Airlines case was not provided or processed for the purpose of AI training, so there would certainly be a change of data use involved, for example. Even in the US, where Federal level data protection legislation seems as far off as ever it was, despite positive noises periodically, how the Google/Spirit case plays out may have regulatory/legal best practice ramifications. Enterprises may find it in their best interests to factor in stronger workforce-specific privacy terms and conditions in future contracts to avoid a similar challenge arising from future potential bankruptcy sales. My take You wrote an email. You didn’t realize you were training an AI model. Or that you were creating a sellable asset. You do now. This is going to be a very important case to keep an eye on as the precedent that it sets is going to be returned to time and again in the coming years. Watch this space. We’ll return to this after 9 September when the court reports back on its thinking. This one matters.