[Anthropic Introduces New Defense Measures, Reduces Model Extraction Attack Data Theft by 90% to 99%] Deep Tide TechFlow reports that on October 13, according to an official disclosure from Anthropic, the company recently deployed a new defense mechanism that has reduced data acquisition from distillation attacks targeting its Claude model by 90% to 99%. Illegal distillation attacks primarily involve using networks of fake accounts to covertly extract advanced model capabilities at low cost, posing risks to cybersecurity and national security. To counter these attacks, Anthropic implemented three core measures: 1. Strictly limiting the flow and access permissions of reasoning processes through mechanisms such as "preserved thinking." 2. Introducing a dual-stage classifier system with an interception rate comparable to biosafety-level defenses, achieving a false positive rate of less than 0.1%. 3. Raising the attack threshold by banning tens of thousands of violating accounts and reinstating charges for intercepted malicious requests. Official data shows that the new measures have reduced the number of stolen reasoning tokens to less than one-tenth to one-hundredth of the original amount. Anthropic calls on industry developers and cloud platforms to collaborate in preventing the illegal extraction of model capabilities.