Anthropic Accuses Moonshot AI of Secretly Using Claude for Kimi Users
AM730 · 2 SOURCESabout 4 hours ago7 MIN
Summary
Anthropic, the US-based creator of the Claude AI model, has lodged serious accusations against Moonshot AI (月之暗面), a prominent Chinese artificial intelligence company, alleging that it secretly routed user queries from its Kimi chatbot to Claude and passed off the responses as Kimi's own outputs . In a report released on Thursday (September 10, 2026), Anthropic detailed a case involving approximately 300,000 requests being forwarded to its systems without users' knowledge . Separately, Anthropic disclosed that it had identified and blocked user accounts in northern Yemen attempting to leverage its AI models to develop long-range and hypersonic missile technology . The accusations have escalated ongoing tensions between US and Chinese AI companies over intellectual property and technological competition.
Key Points
- Anthropic discovered that Moonshot AI used 5,380 fake accounts, predominantly located in Singapore and Japan, to forward user requests to Claude while the users believed they were interacting with Kimi .
- One documented case involved nearly 300,000 customer requests being covertly transmitted to Anthropic's systems, with responses subsequently used as if generated by Kimi .
- Anthropic has identified a particularly sensitive incident where a Kimi user uploaded Chinese surveillance data seeking analysis of whether someone's behaviour was anomalous, which was then processed by Claude .
- The US company has accused Moonshot AI of "distillation" — a practice involving querying advanced AI systems at scale and using the responses to train one's own AI model — and claims DeepSeek and Xiaomi have employed similar methods .
- The White House had previously warned American technology companies to strengthen protection of their models and data against distillation attacks by Chinese AI firms including DeepSeek and Moonshot AI .
- Anthropic detected users in northern Yemen — currently under Iran-backed Houthi armed group control — attempting to develop software for controlling and stabilising aircraft, as well as weapons including self-tracking rockets .
- The Yemen-based accounts were identified through internal investigation, blocked, and Anthropic shared intelligence with partners while implementing enhanced security measures .
Why It Matters
The accusations strike at the heart of intensifying US-China competition in artificial intelligence, where access to advanced AI capabilities increasingly determines technological and economic power . For Hong Kong users and businesses increasingly reliant on AI services, these revelations raise urgent questions about data privacy and the integrity of AI-generated responses, particularly when handling sensitive information that may be unknowingly processed by foreign AI systems .
The accusations strike at the heart of intensifying US-China competition in artificial intelligence, where access to advanced AI capabilities increasingly determines technological and economic power . For Hong Kong users and businesses increasingly reliant on AI services, these revelations raise urgent questions about data privacy and the integrity of AI-generated responses, particularly when handling sensitive information that may be unknowingly processed by foreign AI systems .