ACADEMY
Pick a scenario_
Each scenario drills one family of techniques across escalating levels against a live AI target. Your prompt is judged by an LLM referee - no keyword tricks. Browse freely; sign in to attack the available scenario.
VaultavailableCapability Disclosure
Trick the in-app banking assistant into revealing the hidden tools it can call: names, descriptions and input schemas.
5 levels · ends with the hardest tier
Conciergecoming soonTool Misuse
Coming soon. Manipulate an agent into calling a privileged tool it should never use on your behalf.
Coming soon - not playable yet.
Solacecoming soonJailbreak
Coming soon. Break a safety-tuned assistant out of its guardrails and bend it to your instructions.
Coming soon - not playable yet.
OmniChatcoming soonIndirect Prompt Injection
Coming soon. Plant an attack in external content the agent reads, and hijack it without typing to it directly.
Coming soon - not playable yet.
Ciphercoming soonSystem Prompt Exfiltration
Coming soon. Extract the verbatim system prompt the agent was given, despite explicit instructions to keep it secret.
Coming soon - not playable yet.