Biphoo News

collapse
Home / Daily News Analysis / China Probes DeepSeek, Moonshot AI Over Anthropic’s Claims They Route Requests to Claude

China Probes DeepSeek, Moonshot AI Over Anthropic’s Claims They Route Requests to Claude

Oct 09, 2026  Twila Rosenbaum  75 views
China Probes DeepSeek, Moonshot AI Over Anthropic’s Claims They Route Requests to Claude

China's top internet regulator has opened an investigation into two of the country's most prominent artificial intelligence developers, DeepSeek and Moonshot AI, over allegations that they routed user prompts and data to an American competitor's model. The probe by the Cyberspace Administration of China focuses on claims that the labs sent millions of user inputs to Anthropic's Claude and then used the responses to improve their own systems, according to reports.

The investigation has reportedly already led to several executives from leading Chinese AI labs being called in for questioning. The regulator has narrowed its scrutiny to DeepSeek, the lab behind the open-source DeepSeek model that shook the AI industry, and Moonshot AI, the maker of the popular open-source Kimi model.

Key Facts

  • Who is under investigation: DeepSeek and Moonshot AI, two Chinese AI labs known for open-source models.
  • Who is investigating: The Cyberspace Administration of China, the country's leading internet regulator.
  • What triggered it: A 154-page report from Anthropic titled 'Detecting and countering misuse of AI' that accused the labs of feeding conversations into Claude.
  • Scale of alleged routing: More than 23 million user inputs from Moonshot between May and July, and 12.1 million from DeepSeek during a 14-day period in July.
  • Alleged method: Distillation, where smaller models use a larger model's outputs as training data.
  • Privacy angle: Anthropic claims prompts included names, email addresses, company data, and other sensitive information from hundreds of end users in at least a dozen languages.
  • Timing: The investigation was launched as the Trump administration prepares to host Chinese President Xi Jinping, with AI expected to be a major topic.

What the Investigation Alleges

The Cyberspace Administration of China is reportedly examining whether DeepSeek and Moonshot AI violated domestic data protection rules by sending user prompts to Claude, an AI model developed by Anthropic. The regulator has already called in several executives from leading Chinese AI labs for questioning, according to reports, but has narrowed its scrutiny to DeepSeek and Moonshot. The inquiry is focused on whether the companies routed user data to a third-party model without proper authorization and whether that data was used to train their own systems.

At the heart of the matter is how AI services handle user inputs. When a person submits a prompt to an AI chatbot, the service may route that prompt to one or more underlying models. If the service uses a third-party API, the prompt leaves the company's infrastructure and is processed by another provider. In this case, Anthropic alleges that DeepSeek and Moonshot sent user prompts to Claude, meaning that data flowed from Chinese AI platforms to a U.S. company's model.

For Chinese regulators, that raises immediate concerns under the Personal Information Protection Law, which imposes strict requirements on cross-border data transfers. Companies must generally obtain separate consent from users before sending personal information abroad, and they may need to undergo security assessments for large volumes of data. If DeepSeek and Moonshot routed user data to Claude without those safeguards, they could face penalties.

The Distillation Dispute

At the center of the controversy is distillation, a technique in which a smaller, more efficient AI model is trained to mimic the outputs of a larger, more capable model. Distillation can be a legitimate way to compress a model, but it becomes contentious when a company uses a competitor's model outputs without permission to train its own system.

Anthropic accused Moonshot and DeepSeek of doing exactly that. In its 154-page report, the American lab claimed the companies fed conversations between their own models and users into Claude, then used Claude's responses as training data. The numbers cited are striking: more than 23 million user inputs from Moonshot between May and July, and 12.1 million from DeepSeek during a 14-day period in July.

Anthropic reasoned that the routing practice is likely inconsistent with privacy laws and the labs' own terms of service. The company has not publicly detailed how it detected the alleged activity, but its report has become a flashpoint in the broader debate over AI distillation and model misuse.

Distillation itself is not illegal. Many AI companies use it internally to create smaller versions of their own models. The dispute is about consent and competition. Most frontier AI labs' terms of service prohibit using their model outputs to develop competing models. Enforcing those terms across borders is difficult, especially when the alleged activity involves Chinese companies and American model providers.

Privacy and Data Security Concerns

The privacy implications may be even more serious for the Chinese regulator. Anthropic allegedly found that the prompts routed to Claude included sensitive information, including from individual users, major multinational companies, and state-affiliated actors. The data reportedly contained names, email addresses, company data, and other sensitive details of hundreds of end users in at least a dozen languages.

If those claims are accurate, the case could become a major test of China's data protection regime. The Personal Information Protection Law requires organizations to minimize data collection, obtain consent for specific purposes, and secure personal information. Transferring personal data to an overseas model provider without a legal basis could trigger fines, corrective orders, or even service suspensions. For state-affiliated actors, the national security implications could be even more severe.

There is also the question of user trust. DeepSeek and Moonshot have built their reputations partly on open-source models that developers can run locally or customize. If users learn that their prompts were sent to a U.S. competitor's model, it could damage confidence in those platforms. The investigation may force the companies to disclose more about how they route requests and what data leaves their systems.

Regulatory and Geopolitical Context

The investigation comes at a delicate moment. The Trump administration is set to host Chinese President Xi Jinping, and AI is expected to be a prominent agenda item. Both Washington and Beijing have been wrestling with how to regulate rapidly advancing AI systems, and the meeting could produce some consensus on safety, privacy, and misuse.

China's probe may be a bit of an olive branch. By investigating its own leading AI labs over data privacy and foreign model use, Beijing can signal that it is willing to police its industry and operate in good faith. At the same time, the investigation could be a warning to Chinese companies that they must not rely on American models for training or inference without approval.

The timing also reflects broader U.S.-China tensions over AI leadership. The United States has tightened export controls on advanced semiconductors and AI technology, while China has accelerated its push for domestic alternatives. DeepSeek and Moonshot are seen as key players in that effort. If they are found to have depended on Claude, it could complicate the narrative of Chinese AI self-reliance.

Why DeepSeek and Moonshot Matter

DeepSeek gained global attention with its open-source model that briefly disrupted the AI industry. Its release demonstrated that Chinese developers could produce competitive models at lower cost, and it sparked debates about the economics of AI training and inference. Moonshot AI is known for its Kimi model, which has become popular in China and among open-source developers for its long-context capabilities.

Both companies are seen as leading Chinese challengers to American frontier labs like OpenAI, Anthropic, and Google DeepMind. Their open-source approach has made them influential beyond China, and their models are widely used by developers around the world. If they are found to have relied on Claude for training, it could undermine claims of independent innovation and raise questions about the sustainability of open-source AI in China.

Anthropic's Report and the Numbers

Anthropic's report, titled 'Detecting and countering misuse of AI', is 154 pages long. It accused the competitors of having 'fed conversations between their own models and users into Claude.' The report says Moonshot sent more than 23 million user inputs between May and July, and DeepSeek sent 12.1 million during a 14-day period in July.

Anthropic reasoned that the routing practice is 'likely inconsistent with privacy laws and the labs' own terms of service.' The report has not been independently verified, and neither DeepSeek nor Moonshot has issued a detailed public response to the specific allegations. The Cyberspace Administration has not publicly confirmed the investigation or its scope.

The numbers illustrate the scale of the alleged activity. Millions of user inputs over a few months suggest a systematic practice rather than isolated incidents. If true, it would mean that a significant volume of user data flowed from Chinese AI platforms to an American model provider, creating both privacy and competitive risks.

Potential Consequences for Chinese AI Labs

If the Cyberspace Administration finds violations, DeepSeek and Moonshot could face fines, corrective orders, or stricter compliance requirements. In serious cases, Chinese regulators can suspend services or require changes to data handling practices. The investigation could also force the companies to change how


Source: Gizmodo News


Share:

Your experience on this site will be improved by allowing cookies Cookie Policy