Echo Chamber is a new jailbreak technique that tricks LLMs into generating harmful content by using harmless prompts to slowly shift the conversation. It avoids direct violations, instead poisoning the model’s context with subtle cues. Discovered by NeuralTrust, it works quickly, needs no technical skill, and bypasses safety filters with high success.