AIToday
AI Coding AssistantsAI Safety & AlignmentOpen-Source AIAI WeeklyPublished: Oct 25, 2026, 06:00 JST

Anthropic open-sources Claude Code sandbox as @anthropic-ai/sandbox-runtime

Send on LINE
Anthropic open-sources Claude Code sandbox as @anthropic-ai/sandbox-runtime

3 Key Points

  1. What happened

    Anthropic published srt (sandbox-runtime) on npm under Apache-2.0, wrapping native OS primitives: sandbox-exec with Seatbelt on macOS, bubblewrap on Linux, and a Windows Filtering Platform build with NTFS ACLs.

  2. Why it matters

    Developers running AI agents that execute commands can apply the same filesystem and network restrictions without a container, including allow-only network policies and drop-in sandboxing of MCP servers, though Anthropic calls it a research preview with evolving APIs.

WHO IT HITSDevelopers and platform teams deploying AI coding agents that run shell commands or third-party tools can now apply OS-level filesystem and network restrictions without containers; Windows shops will need to weigh the documented gaps, such as traffic through system services like Windows Update escaping the WFP fence.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The release moves a component that was internal to Claude Code into public view. According to the README, srt is a lightweight tool for enforcing filesystem and network restrictions on arbitrary processes at the OS level, without requiring a container, and it wraps each platform's native primitives rather than building its own isolation layer.

The feature list goes beyond simple file blocking: allow-only network policies with wildcard domain support, Unix-socket allowlisting, experimental TLS termination for MITM inspection of HTTPS, credential masking for environment variables and files, and AWS SigV4 request signing with masked keys. The docs show MCP servers being sandboxed by changing a single command field, from npx to srt.

Anthropic frames the release as a research preview developed for Claude Code to enable safer AI agents, and cautions that APIs and configuration formats may evolve. The Windows backend has documented holes: tools built on the system TLS stack, including System32\curl.exe, PowerShell's Invoke-WebRequest, .NET HttpClient and winget, cannot be used, and traffic routed through system services such as Windows Update escapes the WFP fence.

FAQ
How does srt sandbox a process on each operating system?
On macOS it uses sandbox-exec with dynamically generated Seatbelt profiles. On Linux it uses bubblewrap with network namespace isolation, and on Windows it combines a Windows Filtering Platform egress fence with a write-restricted token and explicit NTFS ACLs.
What does it take to set up the Windows backend?
It ships a bundled srt-win.exe helper for x64 and arm64, and a one-time elevated windows-install step provisions a dedicated srt-sandbox local user account and the machine-wide WFP filter set.
Can I sandbox an MCP server with it?
Yes. The docs show drop-in MCP server sandboxing by swapping the command field in an .mcp.json entry from npx to srt.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleBase hits 138.5 million x402 AI agent payments