China investigates DeepSeek over alleged data transfers to US AI models

China’s cyberspace regulator is investigating DeepSeek and Moonshot AI after Anthropic alleged that Chinese developers redirected user queries through its Claude model, according to The Information, as reported by ForkLog. The immediate significance is that a dispute over model distillation has become a data-governance and national-security matter too.

DeepSeek Watch analysis

What happened

ForkLog reports that China’s Cyberspace Administration summoned representatives from seven companies named in Anthropic’s September report before focusing its investigation on DeepSeek and Moonshot. Officials are interviewing executives and employees, and the regulator has not decided whether sanctions will follow.

Anthropic alleges that Moonshot redirected more than 23 million queries to Claude between May and July, while DeepSeek sent more than 12 million queries over 14 days in July. The claims have not been independently established in the supplied evidence. DeepSeek and Moonshot did not respond to The Information’s requests for comment, according to ForkLog.

The report says some queries may have contained confidential business, government and public-safety information. One cited example alleges that surveillance data from Chengdu, including material connected with monitoring near military and defence research sites, was transmitted through Moonshot’s Kimi. Those are allegations, not findings by the Chinese regulator or a court.

Why it matters

This is no longer only an argument about whether one AI company copied another model’s capabilities. If the reported investigation confirms that sensitive prompts were sent abroad without users’ knowledge, Chinese data-transfer rules and the privacy practices of AI developers could become central to the case.

For users, the practical warning is blunt: a query sent to one AI service may travel through another model or provider if a company is using external systems for training, evaluation or capability extraction. The small print has acquired a passport.

Our read

The investigation is worth reporting because it introduces a concrete state response to a technically messy allegation. But the strongest claims remain attributed to Anthropic and the reporting chain, not presented as settled fact. The next useful evidence would be a CAC finding, a response from DeepSeek or Moonshot, or technical documentation showing what data moved and under whose authorisation.

What to watch

• Whether CAC announces sanctions, formal findings or the scope of any data review.
• Responses from DeepSeek, Moonshot and the other companies named by Anthropic.
• Evidence showing whether queries were knowingly routed to Claude and what they contained.
• Any US-China AI incident or data-governance mechanism discussed alongside the dispute.

Discussion spark: Should AI companies be required to disclose every external model or provider that processes a user’s prompt, even when that disclosure makes experimentation and model development slower?

not affiliated with or endorsed by DeepSeek