AI BriefingTemenos, a Sandbox Dedicated to Agent-Executed Code, Released
Key point
The Temenos project has been released, isolating only the code that AI agents execute rather than the agents themselves, securing both security and convenience at the same time.
Details
Existing AI agent sandboxing methods used to confine the entire agent within a Docker container or VM. This approach causes problems such as loss of host credentials, exposure of API keys, the need to rebuild images, and security vulnerabilities from a shared kernel.
The newly released Temenos takes a structural approach that separates the agent from the executed code.
- Structural separation: The agent operates normally on the host, retaining authentication, updates, API, and MCP.
- gVisor-based isolation: Only the code the agent executes (Shell, scripts, etc.) runs inside a rootless gVisor sandbox.
- Enhanced security: Within the sandbox, the host file system is not visible, and a network-blocking option is provided.
- Enforced execution path: Through structural design rather than prompt instructions, the agent is blocked from directly calling the host's native tools (Bash, Read, Write, etc.), and all execution paths are directed inside the sandbox.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Rootless Linux Sandbox 'Drop' Released for Isolating AI Agents and Third-Party ProgramsWhiteboard Launches Open-Source VS Code Fork for Human-AI Collaborative Software Design
OpenComputer Hosts 'Ship Video' Agent Using Claude Opus 5.5 to Generate Explainer Videos from Code