Anthropic's New Safeguards Cut AI Distillation Attacks by 90-99%
Anthropic announced new measures that reduce data stolen in AI distillation attacks on its Claude models by 90% to 99%. These safeguards limit access to Claude's reasoning, making extraction attempts 10 to 100 times lower. The company also blocked proxy network use and banned tens of thousands of accounts linked to distillation, citing concerns over Chinese startup Z.ai. Anthropic claims its safety system is highly robust with a false positive rate under 0.1%.
ET Tech·


