Anthropic has implemented new methods that significantly reduce the effectiveness of distillation attacks targeting its Claude AI model. The company reports that these defensive measures have decreased the amount of training data collected for distillation models by approximately 90% to 99%. The development marks a major advancement in protecting proprietary AI systems from unauthorized replication through systematic querying and data extraction techniques.