By analyzing unsuccessful jailbreak attempts, developers can train the model to recognize and reject similar prompts in the future.
A jailbreak prompt is a specific input designed to bypass safety filters and content guidelines in large language models (LLMs) such as those in the Gemini family of models Gemini Jailbreak Prompt