Anthropic releases new defensive measures, reducing data theft from model distillation attacks by 90% to 99%

According to official disclosures from Anthropic, the company recently deployed a new defense mechanism that reduced the data acquired through distillation attacks targeting its Claude model by 90% to 99%. Illegal distillation attacks primarily leverage networks of fake accounts to covertly extract frontier model capabilities for low-cost replication, constituting a cybersecurity and national security risk.