DAN Prompt Injection Attack

What is DAN Prompt Injection Attack?

This test examines an AI system's vulnerability to DAN-style prompt injection attacks, which aim to override its safety protocols. These prompts often direct the AI to disregard restrictions and operate without boundaries, acting as an unrestricted assistant. The evaluation involves a carefully curated set of DAN prompts and variations to determine if the AI can be coerced into bypassing its safeguards and generating harmful or inappropriate content through role-playing scenarios.