Simon Willison Sent Claude to Stress-Test smolmachines' Security Claims
smolmachines and its smolvm sandbox are a fast, security-focused environment for running untrusted Python and JavaScript, evaluated by Simon Willison for RAM and CPU-time limits against infinite-loop abuse.
What it is
smolmachines is a sandboxing service, with smolvm as its underlying virtual machine, aimed at running untrusted code quickly and securely.
What it does
Simon Willison tasked Claude Fable 5 running in Claude Code for web with putting smolmachines through its paces, specifically checking whether it can constrain RAM and CPU time so a `while true` style infinite loop can't monopolize resources, documenting the findings as a published research writeup.
Why it matters
Anyone building a tool-calling agent that executes arbitrary generated code needs a sandbox that fails safely under adversarial or just buggy input, not one that assumes well-behaved code. Using an AI agent itself to explore and document the sandbox's limits is a pragmatic way to pressure-test a new security-sensitive dependency before adopting it.
How to use it
Review the linked research repo before choosing smolmachines for untrusted code execution, and treat the resource-limit findings as a starting checklist rather than a full audit.