**[Anthropic Strengthens Defense Against Distillation Attacks, Tens of Thousands of Related Accounts Banned]** Beating AI News Flash: Anthropic has announced that preliminary evidence indicates its new protective measures have reduced the number of Claude inference tokens extracted by distillation attackers from its platform by approximately 90% to 99%. The company stated that attackers typically use networks of fake accounts to batch retrieve model responses, unauthorizedly replicating its capabilities. Among these, complete inference records are more valuable for training purposes than final answers. The new measures include restricting the transfer of inference records across conversations, models, and accounts, deploying a two-stage classifier to identify and block extraction requests, and throttling or banning accounts that repeatedly attempt extraction. Anthropic noted that the classifier's false positive rate is below 0.1%, and tens of thousands of accounts related to distillation have been banned, often within two hours of the attack's initiation. Additionally, starting September 24, the company resumed charging fees for requests intercepted by security measures in specific categories like distillation. Anthropic acknowledged that some developers may need to update their integration methods, and the classifier may occasionally block legitimate requests. It remains unclear how long the protective measures will remain effective, as the company has yet to encounter highly sophisticated and persistent bypass attempts. Continuous iteration of defense mechanisms is still necessary. [Original Link]