Anthropic's 30-day chat retention rule set off a race with OpenAI to keep less customer data

Fortune reports that data privacy has become a key consideration for companies choosing an AI vendor ^[1]. The evidence that Anthropic's June rule on Fable and Mythos ^[2] cost it customers is mostly anecdotal, though. A source who advises clients on becoming AI-native, and who spoke anonymously because his firm works closely with Anthropic, said many companies had an "almost allergic reaction" ^[3]. "The industry at large totally rejected it, and was pretty angry about it," he said ^[4]. The reporting does not include a figure for contracts lost or spending moved.

The labs' own behaviour is firmer evidence. OpenAI saw an opening after the announcement and used an Aug 19 blog post to restate full zero data retention for enterprise customers, while previewing Private Safety Processing, which lets customers keep data in their own cloud ^[13]. Anthropic answered on Sept 1 with a similar zero-retention version that stores data in the customer's own cloud account ^[14]. That came 13 days after OpenAI's post ^[1] and less than three months after the mandate ^[2].

Satya Nadella framed the July warning as a pricing problem. "You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful," the Microsoft chief executive said ^[7]. Customer-cloud storage, now offered or previewed by both labs, is an attempt to remove the second of those payments ^[13][14].

The sovereign alternative means the buyer controls its own stack, which Fortune describes as owning the chips, setting up proprietary cloud infrastructure and running open-source models in its own environment ^[10]. The term used to belong to governments and spread into corporate use this year ^[18]. "Data sovereignty is something we're talking a lot about," said Brooke Hopkins, founder of voice AI company Coval ^[17]. Each of those layers is something the company would own and staff. "You're trading control for responsibility, right?" said Rob Gregory, chief information security officer at Optiv. "Because there is a significant amount of responsibility when you bring that internally." ^[12]

The record fits more than one outcome. The own-cloud terms may be enough, so sovereign AI stays a conversation and the labs keep the accounts. Open models may close the gap in coding and financial analysis, and companies would then move for capability while citing privacy ^[11]. Or the spending may go through hosts such as Amazon Bedrock, which Fortune calls another common route to the labs' leading models ^[15]. I think the first is likeliest for now, because the labs matched each other's terms within two weeks ^[1] and the in-house option costs staff and, in some areas, performance ^[11]. The case against is trust. Fortune reports that recent "rogue AI" incidents have heightened fears that models will scrape or publish business data without permission ^[16], and a storage clause does not address that.

The view fails if restrictions like Booz Allen's ^[5] keep appearing after Sept 1. That would show buyers no longer take a lab's word on retention, whatever the contract says.