Anthropic Blocks AI Distillation Attacks by Chinese Startup Z.ai
Anthropic implemented new safeguards that reduced data harvesting from AI distillation attacks by up to 99%, specifically naming Chinese startup Z.ai as an attacker.
AI company Anthropic announced on Monday that new safeguards have reduced data harvested from AI distillation attacks by 90% to 99%. These measures restrict access to the reasoning tokens of Claude AI models, preventing rivals from using proxy networks to probe protected reasoning and reproduce advanced capabilities at a lower cost.
Anthropic reported that the number of reasoning tokens extracted by distillers is now 10 to 100 times lower than previous levels. As part of this effort, the company banned tens of thousands of accounts linked to these campaigns. The company specifically identified Chinese AI startup Z.ai, also known as Zhipu, as an entity that attempted to distill capabilities from Claude to develop its GLM-5.3 model.
Anthropic characterized these distillation practices as risks to safety and national security. The company stated that its new safety system maintains a false positive rate of less than 0.1%.
