In this post I have listed some of the sandboxing options available out there to isolate your AI agents from unauthorised access into your system and from exposure of sensitive data and files.
GitHub - anthropic-experimental/sandbox-runtime: A lightweight sandboxing… | Mariam Reba Alexander
Allowing an AI agent "write access" to your local machine is like giving your car keys to a stranger. 👇 Sharing here some sandboxing options I found from online research for developers to isolate AI agents from exposing sensitive data on your system: 1. VS Code Dev Containers - https://lnkd.in/e-hs3E_m A ready-to-use sandbox for local development that works with Docker or in GitHub Codespaces. Prerequisites: Docker Desktop or Docker Engine, Visual Studio Code with the "Dev Containers" extension (ms-vscode-remote.remote-containers), or use GitHub Codespaces 2. Daytona.io - https://www.daytona.io/ open-source, elastic infrastructure for running AI-generated code, providing isolated sandbox environments that you can manage using the Daytona SDK to run and control code execution. The Daytona SDK supports Python and TypeScript interfaces. 3. E2B (Firecracker MicroVMs) - https://lnkd.in/e-vCVQUr Ideal for advanced agent workflows where the AI installs third-party libraries or analyzes untrusted files. 4. Code sandbox - https://codesandbox.io/ 5. Docker sandbox - https://lnkd.in/eRdc_h9c 6. Together code sandbox - https://lnkd.in/eyFiFvvq 7. Anthropic Sandbox runtime (srt) - https://lnkd.in/e4X8W2vs What sandboxing options did you find for coding? #aisecurity #ai #aiagents #securecoding #aisandboxing Note: Always review them before using.
linkedin.com
Top comments (0)