Anthropic Deploys New Defenses Cutting Claude Distillation Attack Effectiveness by 90-99%

Anthropic announced Monday that new safeguards against model distillation attacks have reduced the data extracted from Claude by 90% to 99%, a major escalation in the industrial-scale competition to replicate frontier AI capabilities. The company deployed three defensive mechanisms—restricting reasoning-process access, deploying a two-stage classifier with a false positive rate below 0.1%, and enforcing stricter rate limits and account bans—that have cut "the number of reasoning tokens extracted by distillers" to "10 to 100 times lower than before." Tens of thousands of accounts linked to distillation campaigns have been banned, often within hours of detection.