Safety & Ethics
OpenAI details its defenses against adversarial distillation, cutting reasoning-extraction attack success from 54.0% to 0.2% on GPT-6.2 Astra
On November 20, 2026 OpenAI described how it detects and disrupts adversarial distillation, tying protected reasoning to conversation history in GPT-6.2 Astra and Sol. It reports that adaptive extraction attacks now succ…