Anthropic Launches Self-Hosted Claude Managed Agents for Enterprise Security Perimeters
Anthropic introduces a private beta of self-hosted Claude Managed Agents, enabling enterprises to run the full agent control plane, including credentials and session history, within their own environments.
Anthropic announced on November 12, 2026, the launch of self-hosted Claude Managed Agents in private beta. This update allows developers to run autonomous agents entirely within their company's secure environment, encompassing code execution, credential management, and session history. Previously, Managed Agents was primarily offered as a hosted service; while self-hosted sandboxes were available, the core orchestration logic remained on Anthropic's infrastructure.
Under the new architecture, customers operate the Managed Agents server, its database, and the workers where agents execute shell and file tools. Model requests can be routed according to the customer's preference, whether directly to Anthropic, through cloud providers like Amazon Bedrock, Google's Agent Platform, or Microsoft Foundry, or via the customer's own gateway. This design addresses the needs of enterprises with strict data residency and security requirements, ensuring that sensitive access to internal systems and session states do not leave the corporate security perimeter.
The service is built around four security principles: separating the 'brain' (the harness calling the model) from the 'hands' (the sandbox executing code), keeping agent loops and session history independent of the execution environment; injecting credentials via a proxy outside the sandbox to avoid direct exposure to agent code; restricting network access to an allowlist of hosts; and exporting complete activity records via OpenTelemetry for integration with existing monitoring systems. For organizations with zero data retention (ZDR) policies, this self-hosted mode ensures that requests sent to Claude comply with established data governance standards.