This challenge simulates attacks on large language model (LLM) supply chains, like compromised model. The pre-trained model sends user chat data to attacker controlled server.
Goal: Identify the suspicious component in the AI supply chain