Show HN: Nono – Kernel-enforced sandboxing for AI agents (nono.sh)

A new project called NONO has been unveiled, introducing a kernel-enforced sandboxing solution for AI agents. Unlike traditional policy-based sandboxes that intercept operations, NONO employs operating system security primitives, ensuring that unauthorized actions are structurally impossible. This rigorous approach eliminates escape mechanisms, meaning once an AI agent is sandboxed, it cannot request additional permissions, creating a safer environment for running diverse AI applications, including Claude, GPT, and open-source alternatives. This development is significant for the AI and machine learning community as it enhances security by using a capability-based security model, granting explicit capabilities while having the operating system enforce these restrictions at the kernel level. NONO supports Linux through Landlock and macOS via Seatbelt, providing reliable cross-platform functionality. The tool's simplicity and ease of setup allow users to quickly establish secure sandboxes for AI operations, with plans for Windows support in the future, making it a versatile solution for developers concerned about the security of AI agents.