Skip to content

AI Tools in the Lab

We support the use of AI coding and agentic tools — Claude Cowork, OpenAI Codex, Google Antigravity, GitHub Copilot and similar — for lab work. Explore them, test them, and bring what works to the lab.

Three rules that we operate by make this possible.

  • Synthetic or public data only

    No lab, Institute, or collaborator data goes near an AI tool. An AI model can only touch synthetic or public data; you run code on the real data yourself.

  • Review before it runs

    All code is read by a human before the model is permitted to execute it. No auto-approve.

  • You own the output

    "AI wrote it" is not a defence. If you didn't write it, you review it. You are responsible for all code produced.

Pages
AI Tool Use Policy The full policy. Read this before using any of these tools for lab work.
Quick Reference The one-page version, with checklists. Pin it somewhere.

If you're new

Read the policy once, properly. Then keep the quick reference to hand.

If something in here is stopping you doing good work, say so. The rules exist to let us use these tools safely, not to stop us using them.