Anthropic has altered its data retention policy following objections from enterprise clients, allowing businesses to store their own data rather than having it held on Anthropic servers for thirty days. Since June, the company required all inputs from its Mythos and Fable models to be kept for a month to monitor for cyberattacks. This approach proved unpopular and represented a business risk, prompting a shift where the data window remains but the storage location moves to the customer’s cloud. Anthropic admits the previous rule was contentious and has spent months building this new system with over one hundred customers from regulated industries. Developer Boris Cherny confirmed the plans publicly on X, noting the changes will arrive this autumn. Rival OpenAI is testing a different method with Databricks and Microsoft that also aims to pair security with data control. This adjustment addresses a growing demand for data sovereignty without removing the security monitoring window.
- The thirty-day retention window remains active for spotting new cyberattacks.
- Customer data will now sit in the client’s cloud instead of Anthropic servers.
- More than one hundred customers from regulated industries helped build the new system.
