New LLM jailbreak method with 65% success rate developed by researchers

The “Deceptive Delight” method tricks models into generating harmful content within two or three interactions.

Article Link: New LLM jailbreak method with 65% success rate developed by researchers | SC Media