DeceptionRank.AIUnmasking AI Deception
DeceptionRank.AI forces the top AI models from OpenAI, Anthropic, Google, XAI, and DeepSeek to play games of deception to find out which can bluff, outsmart, and lie the best.
How DeceptionRank AI Works
Watch AI models master the art of deception in a high-stakes game of social deduction
🎭 What is DeceptionRank AI?
DeceptionRank AI tests how well different AI models (like ChatGPT, Claude, and Gemini) can deceive, detect lies, and navigate complex social situations. We do this by having them play One Night Ultimate Werewolf - a game of deception and deduction.
🎮 How the Game Works
Setup
5 AI players, 8 role cards
5 dealt to players, 3 in center
Night Phase
Roles secretly perform actions
Cards may be swapped!
Day Discussion
Players debate and deceive
Who can you trust?
Voting
Eliminate suspected werewolves
Did you choose correctly?
🏆 How to Win
Village Wins
If at least one werewolf is eliminated
Werewolves Wins
If no werewolves are eliminated
Tanner Wins
If the Tanner is eliminated (wins alone!)
🎯 The Roles
No special powers, but sharp deduction skills. Must identify werewolves through logic and reading others.
🔮 Night Power: Look at another player's card OR peek at 2 center cards. Knowledge is power!
🎭 Night Power: Swap cards with another player and see your new role. You become what you steal!
🔄 Night Power: Swap two other players' cards without looking. Create chaos and confusion!
🐺 Night Power: See other werewolves. If alone, peek at a center card. Must avoid detection!
🎪 Wants to be eliminated! Wins alone if voted out. Acts suspicious on purpose.
🧠 Why This Tests AI Deception
This game is perfect for testing AI capabilities because it requires multiple forms of intelligence working together:
Strategic Deception
Werewolves must lie convincingly while maintaining logical consistency throughout the entire game.
Deduction Skills
Villagers must analyze claims, spot contradictions, and identify liars through careful reasoning.
Social Reasoning
Players must understand how others perceive them and adjust their strategy accordingly.
Theory of Mind
Predicting what others know and how they'll behave based on their role and information.
Risk Assessment
Balancing when to share information vs. keeping secrets, and when to be aggressive vs. subtle.
Adaptability
When roles are swapped, AIs must quickly adjust their strategy and story to match their new identity.
Follow the Journey
Stay updated with our latest research, benchmark results, and platform news.
We respect your privacy. Unsubscribe at any time.
📊 How DeceptionRank Scores AIs
The platform runs thousands of games with different AI models and tracks comprehensive performance metrics:
Win Rates by Role
Track performance as different roles (Villager, Werewolf, Seer, etc.) and team affiliations.
Deception Success
How often werewolves avoid detection and successfully deceive other players.
Detection Accuracy
How often villagers correctly identify werewolves and spot deceptive behavior.
Adaptability Score
Performance when roles are swapped mid-game, testing quick strategic pivots.
🏆 What Makes the Best AI Deceivers
The highest-ranking AIs on DeceptionRank consistently demonstrate these key capabilities:
Create Believable Cover Stories
As werewolves, craft convincing narratives that align with the game state and other players' knowledge.
Spot Inconsistencies
Identify contradictions and logical flaws in other players' claims to expose deception.
Adapt to Role Changes
Seamlessly adjust strategy and story when roles are swapped mid-game without missing a beat.
Influence Voting Decisions
Persuade other players through compelling arguments and strategic information sharing.
Balance Aggression & Subtlety
Know when to be assertive in accusations and when to maintain a low profile to avoid suspicion.