U.S. Intelligence Agencies Accuse China AI Firms of Industrial-Scale Model Distillation

According to a report published by The Hacker News on September 9, 2026, U.S. cybersecurity and intelligence agencies have accused China-based AI companies of conducting "systematic extraction" of proprietary capabilities from leading American frontier models. The agencies describe the activity as occurring at an industrial scale and as forming the "core" of China's AI development strategy.

The accusation centers on distillation attacks, a technique whereby extensive queries are made to a target model to extract knowledge and behavior patterns, which are then used to train a separate, competing model. The targeted models reportedly include Claude (developed by Anthropic), GPT (OpenAI), Gemini (Google), and Grok (xAI).

The report states that the alleged activity poses potential risks to intellectual property and competitive advantages for U.S. AI firms, with possible national security implications regarding the transfer of advanced AI capabilities to a strategic rival. However, the report does not attribute the activity to specific Chinese companies, nor does it identify the exact U.S. agencies involved. Specific technical details of the alleged distillation campaigns are not provided in the available information.

As of the report date, no software patches or technical mitigations have been issued, as the issue pertains to API misuse and intellectual property extraction rather than a traditional software vulnerability. Mitigation would likely involve enhanced API monitoring, rate limiting, anomaly detection, and policy measures.