OpenAI's figure is 0.2%—the success rate of reasoning extraction attacks against GPT-6.2 Astra after the new safeguards were implemented, according to the company. Crypto Briefing reports that these measures protect the model's internal reasoning and are also seen by OpenAI as a step toward promoting industry collaboration and improving AI safety and trust. Attacks such as reasoning extraction target the model's capabilities themselves; bringing the rate down to 0.2% shows that the bar has been raised. My own approach is to first see whether price and volume can sustain themselves. There's no need to rush into chasing a single message.