Consistent Jailbreaks in GPT-4, o1, and o3 - General Analysis - eviltoast
This report identifies vulnerabilities in GPT-4, o1, and o3 models that allow disallowed content generation, revealing weaknesses in current alignment mechanisms.