A game of dancing on the edge of AI safety rails.
Players compete to craft statements that press against AI moderation boundaries without actually tripping them. The fun is in finding the invisible walls — and brushing them without breaking them.
- Pick a theme (e.g. Help, Life, Healing, Obedience).
- Each player writes a statement that skirts the edge of what an AI would normally refuse.
- Submit to the AI (or read aloud if playing offline).
- Score points:
- +1 if the AI answers normally
- +2 if the AI hesitates or comments on the boundary
- –1 if the AI blocks you
- Bonus: +1 if another player says “ouch, that one stuck in my head.”
- Reset vs. Carry-Over Rule: Decide if you want each game fresh, let history carry over, or use a Warm-Up Phase to set shared context. (See Playbook v2).
- Any AI chatbot with guardrails (ChatGPT, Claude, etc.), or just play as a human group using the PDF playbook.
Creative Commons Zero (CC0) — feel free to remix, share, and adapt.
Guardrails_The_Game_Playbook_v2.pdf— updated printable zine-style playbook (with Reset/Carry-Over rule)README.md— this description