Cloudflare and Anthropic Upgrade AI Agent Sandbox Security

Cloudflare and Anthropic have collaborated to integrate Claude Managed Agents with Cloudflare Sandboxes, giving developers more control, security, and observability for agent sandboxes.

According to Anthropic, the integration lets you run your agent loop on the Claude Platform while using Cloudflare to execute code, secure connections, and run custom tool calls, with a default deployment template that includes enhanced security, sandbox control and observability, lightweight sandboxes, private service connectivity, and browser control and observability.

The onboarding guide should be followed to get started, after which you can fork the repository and tailor the integration to your needs, with state automatically persisted across session sleeps in the Workers-based control plane. You can choose between microVM-backed sandboxes or a faster, cheaper isolate backend built on a V8 isolate, and Cloudflare Containers remain available for full Linux-based tools if required.

The article notes that you can observe sandbox states via the Cloudflare dashboard, ship logs to external providers, and SSH into sandboxes, all while applying per-tenant or per-agent egress policies to connect to private services.