Adversarial Distillation: OpenAI Cuts Attack Rate
OpenAI details how it detects adversarial distillation and says attack success against protected reasoning fell from 54.0% to 0.2% in GPT-6.2 Astra.
OpenAI details how it detects adversarial distillation and says attack success against protected reasoning fell from 54.0% to 0.2% in GPT-6.2 Astra.