Cohere CEO Challenged US Claims on Chinese AI Progress
Aidan Gomez argued that Chinese model performance metrics refute assertions of industrial-scale model theft.
Updated on Sept. 19, 2026 in Artificial Intelligence
Live Poll
Should nations prioritize developing sovereign AI systems that operate independently of foreign technology stacks?
Cohere CEO Aidan Gomez challenged US government claims that Chinese AI development relies on the theft of American intellectual property via model distillation. Gomez suggested that the high performance of Chinese models on benchmarks contradicts the theory that they are simply copies of existing technology.
Why it matters
The dispute centers on the trajectory of AI sovereignty, pitting US intelligence assessments of industrial espionage against private sector observations of Chinese technical capability. The outcome of this debate could influence future export controls and domestic AI development strategies.
Chinese GLM models are currently processing 10 trillion tokens per day, utilizing domestically produced chips to circumvent export restrictions. These performance metrics appear to challenge the feasibility of simple model distillation.
The players
Aidan Gomez
CEO of Cohere, a developer of large language models for enterprise applications.
NSA, CISA, and FBI
United States federal agencies responsible for national security and digital intelligence.
China's Ministry of Foreign Affairs
The executive government department responsible for the foreign relations of the People's Republic of China.
The details
Distillation is a technique where a smaller, more efficient model is trained to replicate the output behavior of a larger, more powerful model. Gomez argues that the high performance of Chinese models proves they are not merely distilled copies, as they exhibit capabilities exceeding the original source material. These systems leverage custom domestic hardware to maintain operations while avoiding reliance on restricted international semiconductor supply chains.
Timeline
June 2026: Fraudulent Claude accounts recorded 28.8 million exchanges.
September 8, 2026: US agencies issued a joint advisory on Chinese distillation.
September 9, 2026: China's Ministry of Foreign Affairs denied the accusations.
September 10, 2026: Aidan Gomez discussed the issue on CNBC.
The Tech Race
This development challenges the efficacy of current semiconductor export controls intended to curb Chinese AI parity. It signals a growing friction between official intelligence assessments and the technical realities observed in the global AI market.
Users of AI platforms may see increased security scrutiny as firms tighten defenses against account abuse following the discovery of 25,000 fraudulent Claude accounts. The dispute highlights the difficulty in verifying the origins of competitive AI software in a global market.
The takeaway
The gap between US intelligence claims and private sector performance analysis indicates a shift in how AI technical leadership is measured. Watch for future benchmark reports that might reconcile these differing accounts of Chinese AI efficiency.
Further reading
For more on the current state of the industry, read our latest analysis in Artificial Intelligence.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Should nations prioritize developing sovereign AI systems that operate independently of foreign technology stacks?
