China has reportedly opened investigations into AI companies DeepSeek and Moonshot AI after allegations that the firms secretly routed user requests to Anthropic’s Claude models.
The investigations, reported by The Information, follow accusations from US-based AI company Anthropic that several Chinese AI developers used proxy networks and fraudulent accounts to access Claude and extract its capabilities at scale.
Anthropic said DeepSeek rerouted more than 12.1 million exchanges to Claude over a 14-day period in July. The company alleged that some of these requests came from users who believed they were interacting directly with DeepSeek’s own AI models.
Credits: The Standard (HK)
Moonshot AI Also Accused of Routing User Requests
Anthropic separately accused Moonshot AI, the developer of the Kimi family of AI models, of routing nearly 300,000 customer requests to Claude over a 10-day period.
According to Anthropic, Moonshot used 5,380 fraudulent accounts, with most apparently located in Singapore and Japan. The company alleged that Claude-generated responses were displayed to people who believed they were using Kimi.
Anthropic also claimed that Moonshot captured at least some of the exchanges and used them to extract Claude’s capabilities for training its own models.
Allegations Extend Beyond AI Model Distillation
The accusations involve more than conventional model distillation, a common technique in AI development where outputs from a larger “teacher” model are used to improve a smaller “student” model.
Anthropic characterized the alleged activity as unauthorised access and large-scale harvesting of Claude’s capabilities.
The company said some of the rerouted requests contained potentially sensitive information belonging to Moonshot customers. These allegedly included surveillance-related data, details involving internal corporate systems and credentials.
Anthropic said it could not establish whether Moonshot had informed affected customers that their requests were being sent to a third-party AI model.
DeepSeek Allegedly Used Claude Reasoning Data
Anthropic also alleged that DeepSeek silently redirected selected user requests to Claude. Some of the exchanges reportedly contained sensitive information connected to a Chinese technology company’s AI programme, a Russian government agency and a Chinese municipal police system.
The company further claimed that DeepSeek used Claude’s “thinking signature” to obtain reasoning information that would otherwise have been protected by the model’s safeguards.
According to Anthropic, this technique could allow reasoning information generated by Claude to be collected and subsequently used in AI model training.
Anthropic Identifies Multiple Chinese AI Firms
The DeepSeek and Moonshot allegations form part of a wider campaign that Anthropic says it has identified since February 2026.
The company said it had detected and disrupted distillation campaigns involving seven China-based AI laboratories. It also named Alibaba, Xiaomi, Z.ai and MiniMax among companies it accused of engaging in different forms of unauthorised distillation.
Anthropic said Alibaba alone was responsible for more than 151 million exchanges between May and July 2026.
The allegations have added to existing tensions between Washington and Beijing over access to advanced artificial intelligence technology. US government agencies have separately raised concerns about Chinese companies attempting to obtain capabilities from US-developed AI systems, while Chinese officials have rejected such accusations and described US restrictions as efforts to constrain China’s AI industry.
Data Privacy Raises Additional Concerns
Beyond intellectual-property questions, the allegations have raised concerns about the handling of customer information.
Anthropic said some of the exchanges involved names, email addresses, corporate information and other potentially sensitive data from users across different countries and languages.
The allegations therefore raise a separate question over whether users were aware that their prompts were being processed by a model operated by another company.
Anthropic said it is strengthening measures to detect proxy networks, fraudulent accounts and other attempts to extract Claude’s capabilities. The company said it may require identity verification when its systems identify potential abuse, including unauthorised resale of Claude or access from unsupported countries.
Credits: Quartz
AI Competition Adds to US-China Tensions
The investigations come as Washington and Beijing continue discussions over AI safety and governance. US and Chinese officials have also agreed to continue talks on AI safety and emergency communication protocols, with another meeting expected in Shenzhen within two months.
For the companies involved, the allegations could create scrutiny on two fronts: how AI models are developed and how customer data is handled when third-party systems are used behind the scenes. Anthropic’s claims remain allegations, and the reported Chinese investigations will be important in determining how the companies respond to questions surrounding their use of Claude and the handling of user requests.
