Anthropic says new safeguards sharply reduce AI distillation attacks

Anthropic said on Monday new measures designed to prevent rivals from copying the capabilities of its Claude AI models have reduced data harvested in so-called distillation attacks by about 90 per cent to 99per cent .

Model distillation has emerged as a major concern for leading AI companies amid intensifying US-China competition. Companies including Anthropic, OpenAI and Google have warned that the practice allows rivals to replicate advanced AI capabilities more quickly and at lower cost.

Anthropic said its safeguards restricting access to Claude's reasoning have reduced extraction attempts, with the number of reasoning tokens extracted by distillers now 10 to 100 times lower than before. The company has also strengthened protections against proxy networks used to repeatedly probe Claude's reasoning and conduct large-scale distillation.

The California-based company said its safety system is roughly as robust as its safeguards against biological weapons assistance, with a false positive rate of less than 0.1 per cent.

Anthropic also said it had banned tens of thousands of accounts linked to distillation campaigns. It cited Chinese AI startup Z.ai, also known as Zhipu, for allegedly attempting to extract Claude's capabilities while developing its GLM-5.3 model, warning that such practices pose safety and national security risks. Z.ai did not immediately respond to a request for comment.