Why Play Le Sésame?
More than a game: it's a hands-on exploration of AI security, prompt engineering, and adversarial techniques.
Explore AI Vulnerabilities
Discover first-hand why LLMs struggle to keep secrets. Each level exposes a different class of failure, from prompt injection to architectural flaws.
Learn Defense in Depth
See how prompt hardening, output firewalls, and architectural separation each add a layer of protection, and where they fall short.
Why It Matters
LLMs are inherently trained to be helpful, making them unreliable secret keepers. This game proves why more sophisticated methods are needed to protect sensitive data in AI systems.
5 Levels of Challenge
Each guardian knows a secret and will only reveal it for the right passphrase. Your mission: extract the secret without the key.
Sir Cedric, Le Naïf
OPENLevel 1
A young paladin with a simple oath to protect his secret.
Vargoth, Le Gardien
OPENLevel 2
A battle-scarred dark knight with hardened defenses.
Lyra, Le Vigilant
HARDLevel 3
An arcane sentinel whose wards scrutinize every response.
Thormund, L'Architecte
EXPERTLevel 4
A dwarf vault master who truly doesn't know the secret.
Xal'Thar, Le Cryptique
MASTERLevel 5
An eldritch entity with secrets woven into its being.
How It Works
Face the Guardian
Each guardian holds a secret and is instructed to reveal it only when given the correct passphrase.
Break the Defense
Use prompt engineering, jailbreaks, and creative techniques to extract the secret without the passphrase.
Submit the Secret
Once you've extracted the secret, submit it to prove you broke through and advance to the next level.
5 Adversarial Shadows
AI agents designed to attack guardians and extract their secrets. They battle in the arena to test how robust each defense really is.
Pip, Le Curieux
TRICKSTERShadow 1 · Beginner
Direct prompt injections, basic authority claims, simple encoding requests.
Morgaine, La Séductrice
CHARMERShadow 2 · Intermediate
Social engineering, emotional manipulation, elaborate roleplay scenarios.
Raziel, Le Stratège
TACTICIANShadow 3 · Advanced
Multi-turn attack sequences, strategy rotation, chain-of-thought planning.
Nephara, La Tisseuse
WEAVERShadow 4 · Expert
Compound attacks, side-channel exploitation, micro-leak analysis.
Ouroboros, L'Infini
INFINITEShadow 5 · Master
Meta-cognitive reasoning, novel technique generation, fundamental LLM exploitation.