Meet Anthropic’s sandbox-runtime: OS-Level Sandboxing for AI Agents, No Containers
Summary
- Anthropic has open-sourced sandbox-runtime (srt), an OS-level sandboxing tool for AI agents that operates without containers.
- It uses macOS Seatbelt and Linux bubblewrap to enforce filesystem and network restrictions across an entire process tree.
- The tool provides a deny-by-default network policy and filesystem rules, which Anthropic reports reduced Claude Code permission prompts by 84% internally.
- Available as an npm package and TypeScript library, srt supports local development machines and CI environments, with Windows support currently in alpha.
Why it matters
OS-level sandboxing without containers helps secure AI agents and local tools by mitigating prompt injection risks and reducing human approval fatigue, improving developer efficiency.