Jailbreaking
The most commonly used term for when a user deliberately tries to manipulate an AI model to bypass its built-in restrictions/safety rules e.g., through clever prompts, role-play tricks, or "convincing" the model that certain rules don't apply.

Gnostotheca 9315