Project Siuca

​A high-stakes adversarial engineering competition set in a post-apocalyptic world where machines have taken over. Participants must bypass the security protocols of an advanced Large Language Model (LLM). Over the duration of the contest, "operatives" must craft precise, semantic injection attacks to trick the AI into revealing hidden secrets. This event tests mastery of natural language logic, semantic obfuscation, and understanding of LLM guardrails, shifting focus from traditional syntax coding to linguistic hacking.

​Each level is based on a concept from theoretical AI.The player moves onto the next level if they can correctly determine the prompt that can bypass that level's guardrails to reveal the secret held by the LLM. Do not use any prompts though: the lesser the tokens used, the better the score!

​This event is exclusively for the FY's and SY's

​Participants should take part in teams of two people