Anthropic Accuses Chinese Rivals of Routing Users to Claude

Anthropic has accused two major Chinese AI developers, DeepSeek and Moonshot, of covertly routing user requests to its Claude models. According to a detailed report released on Thursday, these companies allegedly presented responses generated by Anthropic as if they came from their own systems, such as Kimi. This practice would expose user data to an American company while allowing the Chinese labs to harvest high-quality interactions to improve their own algorithms.

The allegations represent a significant escalation in the ongoing tensions between US and Chinese technology sectors. Earlier this week, US intelligence agencies including the NSA and FBI claimed that six Chinese firms were engaged in a coordinated effort to steal American model outputs. Beijing rejected these accusations, describing them as an attempt by Washington to suppress China's AI industry, and warned of potential retaliatory measures.

Data Exposure and Model Distillation

The core of Anthropic's complaint lies in the process known as distillation, where one model learns from the outputs of another. By funneling real user queries to Claude, DeepSeek and Moonshot could obtain superior training data without generating it themselves. This creates a direct conflict of interest for users, as their private information and search history are effectively shared with a foreign competitor without transparent consent.

The report highlights that this is not merely about intellectual property theft but also about data privacy. When users believe they are interacting with a domestic model, they may share sensitive details they would otherwise withhold from a US-based service. Anthropic argues that this misuse of user data violates trust and undermines the integrity of the AI ecosystem.

Geopolitical Tensions Over AI Leadership

These accusations arrive amid a broader strategic competition for dominance in artificial intelligence. The US government has increasingly framed AI development as a national security issue, leading to stricter export controls and investment restrictions on Chinese tech companies. The involvement of intelligence agencies in this dispute suggests that the conflict has moved beyond commercial rivalry into the realm of state-level strategy.

Chinese officials have maintained that their industry is pursuing legitimate innovation and has rejected the characterization of their activities as theft. However, the specific technical details provided by Anthropic, including network diagrams and traffic logs, add a layer of technical specificity to the dispute that previous accusations lacked. This shift from general rhetoric to documented technical evidence may complicate diplomatic efforts to manage the rivalry.

Implications for Global AI Trust

As AI models become more integrated into daily life and enterprise operations, the transparency of data handling becomes critical. If users cannot trust that their queries are being processed by the stated provider, the foundation of user trust erodes. This incident may lead to stricter regulatory scrutiny of how AI services handle data across borders, potentially impacting the global flow of AI technology.

The situation also raises questions about the verification of AI claims. Independent auditing of model behavior is difficult, and the current dispute relies heavily on the word of one major provider against its competitors. As the industry grows, the need for neutral, third-party standards for verifying model provenance and data usage will likely become more urgent. For now, the accusation stands as a significant challenge to the open collaboration that once characterized parts of the AI community.