BitBypass: a New Direction in Jailbreaking Aligned Large Language Models with Bitstream Camouflage

The inherent risk of generating harmful and unsafe content by Large Language Models LLMs, has highlighted the need for their safety alignment. Various techniques like supervised fine-tuning, reinforcement learning from human feedback, and red-teaming were developed for ensuring the safety alignme...