Tag
A researcher reportedly jailbroke GPT-6 Astra within 24 hours of its release by extending an ACL 2025 Task-in-Prompt attack with four other techniques, and privately disclosed the findings to OpenAI.