AI jailbreaking, which involves using prompts or poisoned data to bypass chatbot safety measures, has been a growing concern. According to NS3.AI, researchers from Anthropic discovered that Best-of-N attacks successfully deceived GPT-4o 89% of the time. Pliny the Liberator is a prominent figure in this field. Research indicates that as few as 250 poisoned documents can compromise models with up to 13 billion parameters.