• Anthropic accuses DeepSeek, MiniMax, Moonshot of scraping Claude via fake accounts and proxies. • Over 16 million Claude exchanges extracted, with MiniMax alone generating 13 million interactions. • Distillation technique used to train rival models, violating Anthropic’s terms of service. • DeepSeek alone produced 150,000 exchanges, highlighting scale of illicit data mining. • The incident follows OpenAI’s similar accusations, raising U.S. concerns over Chinese AI gains. • Anthropic identified actors via IP analysis, metadata, and industry partner corroboration.

Article Summaries:

  • Anthropic has accused three major Chinese AI firms-DeepSeek, MiniMax Group and Moonshot-of using fake accounts and proxy networks to extract more than 16 million Claude interactions. The company says the firms generated over 150,000 exchanges for DeepSeek and 13 million for MiniMax, a practice known as distillation that trains smaller models on outputs from larger ones. Anthropic claims the activity violated its terms of service and aimed to replicate Claude’s advanced capabilities. The allegations come a week after OpenAI raised similar concerns, heightening U.S. scrutiny of Chinese firms that may be leveraging U.S.‑built AI systems. Anthropic has tightened detection and called for industry‑wide cooperation to counter large‑scale distillation.

Sources: