‘Bad Likert Judge’ Jailbreak Bypasses Guardrails of OpenAI, Other Top LLMs

A novel technique to stump artificial intelligence (AI) text-based systems increases the likelihood of a successful cyberattack by 60%.