Anthropic Alleges Distillation Campaigns by Alibaba, Moonshot AI, and DeepSeek

It is a technique for training a smaller or cheaper model to mimic the outputs of a larger model, typically by sending large volumes of structured queries and using the responses as training data.

According to the report, Anthropic detected abnormal access patterns from accounts linked to Alibaba, Moonshot AI, and DeepSeek, suggesting organized attempts to extract model capabilities rather than normal usage.

There is no one-size-fits-all answer. Assess legal exposure, data sovereignty, and audit trails. Government and regulated entities may prefer to keep sensitive workloads on local or vetted providers.

Use rate limiting, automated pattern detection, restricted API access, output watermarking or perturbation, and updated terms of service backed by legal enforcement.