Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek

Anthropic has released a report detailing what it describes as persistent model distillation attacks originating from Chinese AI companies, including Alibaba, Moonshot AI, and DeepSeek. Model distillation involves training a smaller, more efficient AI model by using the outputs of a larger, more powerful model. Anthropic alleges that these companies have been systematically querying its models to extract proprietary knowledge and replicate their capabilities, a practice that has intensified alongside growing global competition in the generative AI sector. The report highlights the ongoing tension between AI labs regarding intellectual property and the security of large language models. As the race to achieve state-of-the-art performance accelerates, Anthropic is calling for greater scrutiny and defensive measures against unauthorized data extraction. This development underscores the broader challenges in protecting foundational AI models from competitive exploitation in an increasingly crowded and high-stakes market.