The “Crescendo” attack uses a chain of seemingly benign prompts to achieve an adverse output.
Article Link: Microsoft’s ‘AI Watchdog’ defends against new LLM jailbreak method | SC Media
The “Crescendo” attack uses a chain of seemingly benign prompts to achieve an adverse output.
Article Link: Microsoft’s ‘AI Watchdog’ defends against new LLM jailbreak method | SC Media