Anthropic will allow enterprise customers to store logs of their model requests in their own cloud rather than on the company’s servers. Since June 2026 Anthropic had retained this data for 30 days to detect cyberattacks spread across multiple requests — a requirement that hurt sales to large clients with strict security standards.

Starting this autumn the monthly monitoring window stays, but the data will no longer leave the customer’s perimeter. The architecture took months to build and was tested in a closed beta with roughly a hundred clients.

Related: Microsoft Sovereign Cloud Adds Governance, Productivity, and Support for Large AI Models

Source: Boris Cherny on X