Radar / AI security / Anthropic sandbox-runtime
Anthropic sandbox-runtime: OS-level sandboxing for AI agents, no containers required
Anthropic's sandbox-runtime (srt) is trending with developers this week: an open-source tool that wraps any process in filesystem and network restrictions using native OS primitives — Seatbelt on macOS, bubblewrap on Linux — plus a proxy-based network allowlist, without spinning up a container.
Why it matters
It is the sandbox behind Claude Code's sandboxed bash tool, now usable with any agent, MCP server or plain command — and Anthropic reports it cut Claude Code permission prompts by 84% in internal use.
What you could build with it
A team running coding agents in CI could prefix untrusted agent commands with srt so a compromised or confused agent can read only the working directory and reach only allowlisted hosts, turning agent sandboxing into a one-line change.
Does it hold up?
Installable today via npm and already proven inside Claude Code, but it is a beta research preview with alpha Windows support, so production use on Linux/macOS only.
Built with Anthropic sandbox-runtime
- Meet Anthropic's sandbox-runtime: OS-Level Sandboxing for AI Agents, No ContainersMarkTechPost · Oct 24 writeup of the trending tool: npm package v0.0.81, Apache-2.0, Node 22.12+, and the 84% fewer permission prompts claim.
- @anthropic-ai/sandbox-runtime npm package overviewSocket.dev · Package analysis confirming the OS-level approach: sandbox-exec on macOS, bubblewrap on Linux, proxy-based network filtering.
- List of coding agent sandboxesGitHub Gist · Curated sandbox comparison listing Anthropic's Sandbox Runtime (srt) among local sandboxes built for coding-agent CLIs.
Learn more
First spotted on github: source.
More AI security
iFixAi: independent auditing of AI agentsIndependent AI-agent auditor: 250+ inspections across 69 misalignment categories (red teaming + operational assurance)…security · JEV 0.76Cloudflare security-audit-skill: six-phase agent-driven security audits with independently verified findingsOpen-source MIT coding-agent skill that runs a codebase through six audit phases — reconnaissance, coverage-led…security · JEV 0.74OpenAI, Anthropic and Google DeepMind jointly unveil cyber-focused safety models and safeguardsThe three rival labs disclosed weeks of behind-the-scenes coordination and jointly released a set of cyber-focused AI…security · JEV 0.73OpenAI textGrain watermarking for ChatGPT and Codex in the EUOpenAI will roll out its invisible textGrain watermarking system to ChatGPT/Codex users in the EU over the coming…security · JEV 0.65Cloudflare security-audit-skill: multi-phase security audits for coding agentsA coding-agent skill that runs multi-phase security audits with independently verified, machine-checkable findings.security · JEV 0.63LiveNerf: a pre-registered benchmark for post-release model driftOpen-source, pre-registered 30-day benchmark that detects whether Claude Opus 5.5 quietly gets worse after launch.security · JEV 0.61
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.