Claude Managed Agents add two safety features

Claude托管代理新增两项安全功能

Claude Managed Agents added two safety improvements: self-hosted sandboxes keep agent execution environments in the user’s infrastructure or hosted sandbox provider, while MCP tunnels let agents connect to services inside the user’s security boundary.

Why it matters: HKR-K and HKR-R pass: the post names two agent-safety mechanisms and a concrete execution-boundary change. HKR-H is weak, and this is not a model release, so it sits in the low featured band.