OpenAI reports 15,000 users linked to China's Moonshot trying to copy AI.

OpenAI has disclosed that it disrupted a large-scale operation involving more than 15,000 users who attempted to extract the hidden reasoning processes behind its advanced AI models. The company identified a core cluster of these accounts as being linked to individuals associated with Moonshot AI, the Chinese startup that developed the Kimi chatbot. This revelation highlights an escalating industrial effort to reverse-engineer the internal logic of leading artificial intelligence systems, moving beyond simple data scraping to a more sophisticated extraction of proprietary cognitive frameworks.

The move comes amid growing tensions over intellectual property in the global AI race. For months, tech industry observers have noted that Chinese developers are aggressively benchmarking against American models to close the performance gap. By targeting the "chain of thought" or intermediate reasoning steps, rather than just final answers, attackers can deduce the specific algorithms and training data structures that drive a model’s behavior. OpenAI argues that this specific type of probing violates its terms of service and represents a direct threat to its competitive advantage. The scale of the operation, with thousands of distinct user accounts, suggests a coordinated campaign rather than isolated academic curiosity, pointing to a significant resource commitment by the entities involved.

Moonshot AI, the startup behind the Kimi assistant, has not publicly commented on the specific allegations regarding the 15,000-user cluster. However, the company has previously emphasized its focus on long-context processing and efficiency in its own model development. Critics of OpenAI’s stance often point to the open nature of much of the AI research community, arguing that understanding how models work is essential for safety and transparency. OpenAI, conversely, maintains that its reasoning capabilities are a trade secret developed through years of expensive research and development. The company claims that by disrupting this campaign, it has protected the integrity of its technology and prevented competitors from gaining an unfair shortcut to replicating its most advanced features.

This incident marks a significant shift in how AI companies approach security. It is no longer just about preventing data theft or stopping bot attacks, but about protecting the internal architecture of the models themselves. As the distinction between consumer AI and frontier research blurs, the line between legitimate testing and industrial espionage becomes increasingly difficult to police. Regulatory bodies in both the United States and China are watching closely, as this dispute could influence future export controls and data sovereignty laws. For the broader industry, the message is clear: the next frontier of competition lies not just in computing power, but in the secrecy of the reasoning engine that drives it.