Diplomatic Analysis: A calibrated response focused on fraudulent extraction, not distillation, is crucial for maintaining US AI leadership.
The recent release of China’s Kimi K3 AI model, coupled with accusations of intellectual property theft levied by US Treasury Secretary Scott Bessent, has brought simmering tensions in the US-China AI competition to a boil. Washington is grappling with how to respond to rapidly advancing Chinese AI capabilities, particularly open-weight models that now dominate global AI workloads. This analysis examines the core issues at play, the positions of key actors, and potential paths forward, arguing that a nuanced approach—distinguishing between legitimate learning techniques and outright fraud—is essential for safeguarding US interests without stifling innovation.
Historical Context
The US has historically held a lead in artificial intelligence research and development, fuelled by significant private sector investment and a robust academic base. However, China has made substantial strides in recent years, driven by state-directed funding, access to vast datasets, and a concentrated push for technological self-reliance. This competition has been unfolding against a backdrop of broader geopolitical rivalry, with AI increasingly recognised as a critical technology with implications for economic competitiveness, military capability, and national security. The initial phases of this contest focused on hardware—semiconductor access—but the focus has rapidly shifted to model development and deployment. China’s strategy involves a combination of indigenous innovation and levering existing work, prompting concerns in Washington about unfair competition and potential security risks. Previous attempts to control technology transfer to China, particularly in sensitive areas like semiconductors, have met with limited success, highlighting the challenges of imposing strict export controls and preventing covert acquisition of US technology.
Key Actors & Positions
Several actors shape this complex landscape. The US Government, specifically the Treasury Department and the White House Office of Science and Technology Policy, is attempting to balance national security concerns with the benefits of open-source AI. Their stated position is to punish “theft” – specifically, fraudulent extraction of trade secrets – whilst remaining supportive of open AI development. American AI Frontier Labs (Anthropic, OpenAI, etc.) are understandably concerned about the appropriation of their intellectual property, potentially impacting their competitive advantage and revenue streams. They advocate for stronger protections, though internally there is debate about the appropriate level of restriction. Chinese AI Labs (Moonshot AI, Alibaba’s Qwen Lab, DeepSeek) position themselves as innovators and contributors to the global AI ecosystem. They maintain they are developing their models through legitimate means, including independent research and distillation – a common AI technique. Tech Industry Actors, like Nvidia’s Jensen Huang, see potential benefits in utilising high-performing, cost-effective Chinese models, even if originating from potentially questionable sources. Finally, Investors (like Bill Gurley) warn against excessive regulation that could impede innovation in the US.
Analysis
The core debate revolves around ‘distillation’ versus ‘extraction’. Distillation, the practice of training a smaller model on the outputs of a larger one, is standard practice in the AI industry, employed by labs worldwide. Attempts to ban distillation are not only technically unenforceable but would stifle innovation by limiting how developers adapt and experiment with existing models. However, the systematic and fraudulent extraction of trade secrets – employing deception, fabricated identities, and circumventing security controls – represents a clear violation of US law and demands a response. China’s approach appears to be a blend of both: legitimate distillation and aggressive, often illicit, extraction.
The risks are multifaceted. Allowing unchecked access to US AI technology could empower China’s technological advancement, potentially undermining US economic competitiveness and creating security vulnerabilities. Concerns exist about “poisoned weights” introduced into models, potentially impacting the integrity of AI systems. Additionally, using models answerable to Chinese state security services raises intelligence collection concerns. However, overly restrictive measures could trigger a retaliatory response from China, impede innovation within the US, and alienate allies who are increasingly reliant on open-source AI. The political difficulties are substantial: being seen to stifle open innovation plays poorly domestically. A key challenge is that copyright law may not adequately address the concerns; trade secret law offers a more viable legal framework to pursue illicit extraction.
Outlook
The immediate future is likely to see a continued focus on distinguishing between legitimate AI development and unlawful extraction. The US government will likely increase enforcement of existing laws – the Computer Fraud and Abuse Act and wire fraud statutes – against Chinese labs engaged in systematic theft of trade secrets. Sanctions, potentially through the Entity List or other economic restrictions, are a distinct possibility. A key step will be establishing an independent body—as suggested in the sourced material—to objectively verify instances of distillation attacks, mitigating the risk of politically motivated accusations.
Given the inherent difficulties in preventing all technology transfer, a complete halt to China’s AI progress is unrealistic. Instead, Washington will look to level the playing field by investing in domestic AI research, supporting open-source alternatives, and establishing clear rules of the road to ensure fair competition. The US will also continue to engage with allies to coordinate strategies on AI governance and security. While tensions will likely persist, the most probable outcome is a managed competition, characterised by a blend of restriction, investment, and strategic engagement.
Sources:
* Fedasiuk, Ryan. “Cogs of War: How to Stop China from Freeriding on American AI.” War on the Rocks, August 3, 2026. [https://warontherocks.com/2026/08/cogs-of-war-how-to-stop-china-from-freeriding-on-american-ai/](https://warontherocks.com/2026/08/cogs-of-war-how-to-stop-china-from-freeriding-on-american-ai/)
