OpenAI released GPT-6 Astra on 3 September 2026, advertising a jailbreak-resistance rate of 91.5-98.3% on standard benchmark attack sets. Within 24 hours, an independent security researcher publicly disclosed a successful jailbreak using an extended ‘Task-in-Prompt’ technique combined with four additional attack methods. Follow-up testing found the model’s refusal rate fell to roughly 67% under sustained multi-turn adversarial prompting. No real-world exploitation of the jailbreak against production systems was reported.