Anthropic disclosed that Alibaba, Moonshot, and DeepSeek conducted large-scale unauthorized distillation campaigns using Claude outputs between December 2025 and August 2026. The activity involved over 183 million total exchanges, with Alibaba alone accounting for more than 151 million interactions, effectively extracting capabilities from a frontier model to improve domestic competitors without authorization. This constitutes a material breach of service terms and raises significant questions regarding data provenance and training data cleanliness in the affected models. For financial institutions, this event introduces specific legal and reputational risks tied to enterprise contracts involving these vendors and highlights the structural vulnerability of proprietary model IP to illicit extraction. Also reported by: https://www.washingtonpost.com/business/2026/09/09/us-china-ai-models-anthropic-trump/9f920ab4-ac33-11f1-b498-8697f35a6743_story.html
342 other tracked developments touching the same claims.