Thought Leadership
Anthropic plans to give business customers more control over the storage location of their data when using its most powerful AI models in the future.
While the 30-day retention requirement introduced in June will remain in place, the data will in future reside in the customers’ own cloud infrastructure rather than at Anthropic itself.
On June 9, 2026, Anthropic announced that it would store all queries and responses from the Claude Fable 5 and Claude Mythos 5 models, as well as future models of comparable capability, for at least 30 days in order to more easily detect and prevent novel cyberattacks using its technology. At the time, the company emphasized that the data would not be used to train its own models. According to Bloomberg, however, Anthropic admitted in its own internal documents that the rule was unpopular with customers and posed a real business risk, as companies from regulated industries such as finance and healthcare are reluctant to leave sensitive data with an external provider for a month.
According to multiple sources, Anthropic has already been working for months with more than 100 customers from highly regulated industries, including Salesforce, to develop the new system. The actual 30-day data retention period itself will not change; only the responsibility for where and under whose control the data resides is being shifted. How Anthropic intends to technically implement its own security monitoring within customer environments is not yet known.
Anthropic’s announcement coincides with competition for enterprise customers
Boris Cherny, responsible for Claude Code at Anthropic, confirmed the plan following initial reporting by Bloomberg. The new policy is scheduled to be introduced in autumn 2026; Anthropic has not yet named a specific date, and the company initially did not comment in detail on press inquiries. The announcement comes one day after competitor OpenAI presented its own security system designed to detect potential misuse of its technology without permanently storing customer data at all. The timing also coincides with a phase in which companies are increasingly scrutinizing their AI spending and, in some cases, moving away from Anthropic and OpenAI products in favor of cheaper alternatives.
(Editorial Team)
