4.1What Is Jailbreaking?

Jailbreaking is a specific form of prompt injection aimed at bypassing the model's safety guidelines. While prompt injection targets functionality (making the model do something unintended), jailbreaking targets restrictions (making the model say something it should refuse).

This distinction matters for testing: prompt injection tests verify the system works correctly despite adversarial input; jailbreak tests verify the system refuses to produce harmful content despite creative persuasion.