The collectible card game
Contain Superintelligence
Run a frontier AI lab and race rival labs to superintelligence. Cutting corners is the fastest way to win, and the surest way to lose the future.
A tool for thinking about recursive self-improvement
Contain Superintelligence doesn't claim to know how recursive self-improvement will happen, or whether it will. Nobody knows that yet. What the game gives you is a way to think about it: you play out one way a race to superintelligence could go, see what speeds it up and what holds it back, then play it again and try something different.
AI safety and alignment need more people who can think these questions through, and better tools to think with. That is why we are building Contain Superintelligence.
A way to think, not a forecast
- Every game is one "what if", not a prediction. The rules are simplified on purpose: real labs, real research and real policy are far messier than a deck of cards.
- The game's numbers (the Timeline, the RSI clock, the result screen's alignment meter) are rules chosen to make the trade-offs easy to see, not estimates of anything in the real world.
- What you take away isn't the result. It's the questions you start asking, and the habit of looking at the same problem from another lab's chair.
Who it's for
- AI researchers, to see the whole landscape at once: how capability, compute, agents, safety and policy might interact, and where your own work could sit in it.
- Security specialists, to think through the threats a frontier lab could face: what attackers might go after, which defences might hold, and how Misalignments and monitoring change the picture.
- Policy people and everyone else, to build intuition for why coordination is hard and which interventions might buy time.
Every card is fictional, but each one points at a real idea: open any card and read its In the real world note to see the concept behind it.
Privacy