Anthropic Introduces Self-Hosted Sandboxes for Claude Managed Agents

Anthropic recently announced the public beta of its self-hosted sandbox feature for Claude Managed Agents, marking a significant step towards enabling enterprises to integrate advanced AI capabilities with enhanced security and control. This new offering allows organizations to execute the tool calls made by Claude Managed Agents directly within their own infrastructure perimeter. This means that sensitive code and data processing can occur within environments fully controlled by the customer, including their existing containers, virtual machines, or serverless functions. The core benefit of this approach is the ability for enterprises to meet stringent data privacy, security, and compliance requirements by keeping the execution of agent-driven tasks within their trusted network boundaries. Previously, the execution of such tasks might have been more tightly coupled with Anthropic's cloud infrastructure. With self-hosted sandboxes, the orchestration layer for the Managed Agents remains with Anthropic, ensuring the intelligent coordination and management of AI workflows. However, the actual computational surface where tools are invoked and data is manipulated is shifted to the customer's chosen environment. This architectural separation provides a powerful hybrid model, combining the intelligence and management of Anthropic's AI with the security and governance of the enterprise's own infrastructure. The public beta, which was unveiled at the "Code with Claude London" event on May 19, 2026, requires a specific beta header for all Managed Agents API calls, indicating its ongoing development phase. The implementation involves a worker component that claims items from a session queue, spawns an execution context—which can be a container, VM, or serverless function—and then downloads the agent's skills to a designated workspace. Tool calls are then run locally within this isolated environment, with results posted back to the Managed Agents queue. This design ensures that while the agent's intelligence is leveraged, the sensitive operational aspects are managed under the enterprise's direct oversight. This capability is particularly relevant for organizations dealing with confidential data or operating in highly regulated industries, offering a robust solution for deploying AI agents responsibly.