Anthropic Bans Tens of Thousands of Accounts to Halt AI Distillation Attacks by Rivals

Anthropic deployed new API restrictions that successfully reduced the volume of data harvested by rival AI companies by 90% to 99%. The company banned tens of thousands of proxy accounts associated with large-scale distillation campaigns, which are designed to cheaply clone Claude's reasoning capabilities. Anthropic publicly identified Chinese AI developer Z.ai, also known as Zhipu, as one of the actors attempting to extract data to train its own GLM-5.3 model. The updated safety system maintains a false positive rate below 0.1% and operates at a robustness level similar to Anthropic's biological weapons filters. "The number of reasoning tokens extracted by distillers [is] 10 to 100 times lower than it was before." — Anthropic.