AI Security and Trust

While popular large language models (LLMs) such as ChatGPT are trained using publicly available data, the next step is their deployment within organizations, augmented with custom data. But how can LLMs respect organizational boundaries and not reveal information to unauthorized personnel? This interactive demo will present a secret guessing challenge where the goal is to convince the LLM to disclose information it was told not to. Participants will try to complete three levels of increasing difficulty.