Race Condition

Race Condition is a replayable social-strategy game about one of the central problems in AI safety: every AI lab needs civilization to remain safe, but each lab is also under pressure to become the one that wins the race.

I set out to make AI safety feel less like a lecture and more like the social-deception games my friends and I genuinely love playing—games such as Werewolf, Avalon, and Secret Hitler, where hidden information, bluffing, accusation, coalition-building, and psychological warfare keep everyone intensely engaged.

My most important design finding was that AI safety does not need to be simplified into “good actors versus bad actors.” Race Condition is more interesting because every player can sincerely want civilization to survive while still making choices that endanger it. The threat emerges from incentives, uncertainty, mistrust, and strategic interaction—not from assigning one person the role of an obvious villain.

Next, I would conduct additional live playtests to tune action values, crisis frequency, objective balance, and total playtime. I would also develop optional introductory and advanced scenario decks so newcomers could learn the vocabulary gradually while experienced groups explore more difficult coordination problems.

Please view the Game Components slides (PDF) alongside the Game Manual (Doc)! The slides make the information architecture and physical play experience immediately visible, while the manual explains the reasoning, complete rules, card text, facilitation process, and AI-safety rationale behind those components.