Anthropic reports cutting distillation attacks on Claude by 10 to 100 times

AIAnthropic says it has made significant progress over the last few months in reducing the effectiveness of distillation attacks against Claude. The company says the methods are explained in a linked research post. The post itself gives no figures or details of the defenses.

AIAnthropic says its new defenses have cut the reasoning tokens distillers can extract from Claude by a factor of 10 to 100 on its own platform. The company describes three measures: controlling where reasoning can go, a two-stage classifier that blocks extraction attempts with a false positive rate below 0.1%, and stricter limits and bans for repeated attempts. Anthropic also names Zhipu's GLM-5.3 and Moonshot's Kimi as cases where distilled capabilities raised cyber and national security concerns.