According to The Defiant, in a single-round cybersecurity test targeting the pre-release GPT-6.2 Astra, a researcher used approximately half the number of queries to develop a jailbreak method compared to the tests against GPT-6 Astra. The report noted that OpenAI subsequently modified its safety guards, but as of the British testers' briefing on November 17, these latest modifications had not yet been independently re-tested by them. This comparison is limited to the aforementioned tests and does not directly represent the current protection efficacy of the live version.