US Authorities Target DeepSeek and Moonshot AI for 'Malicious Distillation' of US AI Models

U.S. intelligence and security authorities have stated that six Chinese artificial intelligence (AI) companies, including DeepSeek and Moonshot AI, extracted large-scale functionality from U.S. AI models for their own model development. The National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Cybersecurity and Infrastructure Security Agency (CISA) revealed in a joint cybersecurity advisory published on August 8 that DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI extracted hundreds of millions of requests from U.S. AI models between late 2024 and the present, amassing billions of tokens.
The authorities noted the activity may have occurred under the direction of the Chinese government. They stated that DeepSeek targeted inference capabilities, professional optimization, and specific domain features for developing its R1 and V3 models, while Alibaba utilized industrial-scale distillation to improve its Qwen model. Distillation involves training smaller models using outputs from larger ones, which authorities claim systematically extracted exclusive U.S. AI company functionalities.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 미 당국, 딥시크·문샷 AI ‘악의적 증류’ 지목