Growing tensions between the United States and China over artificial intelligence are raising concerns that international cooperation on AI safety could be undermined at a critical moment. As increasingly capable AI systems emerge, researchers and policymakers warn that geopolitical rivalry may hinder the development of common safeguards designed to reduce security risks associated with advanced models.
US allegations against Chinese AI firms raise diplomatic stakes
The latest dispute follows accusations by US officials that Chinese artificial intelligence company Moonshot improperly used outputs from Anthropic’s advanced Fable 5 model to develop its own Kimi K3 system. US Treasury Secretary Scott Bessent has indicated that sanctions against the company are being considered.
The allegations centre on a technique known as “distillation”, in which developers train a new AI model using responses generated by a more advanced system. The method can significantly reduce the computing costs required to produce competitive AI models.
At the same time, the US Department of Commerce’s Bureau of Industry and Security is investigating whether Chinese firms, including Moonshot, may have obtained access to advanced American semiconductor chips in breach of existing export controls. Moonshot has not publicly responded to the allegations.
These developments come amid a broader technology dispute between Washington and Beijing, with artificial intelligence increasingly viewed as both a strategic economic asset and a matter of national security.
AI safety dialogue at risk amid growing geopolitical rivalry
Analysts warn planned talks could be disrupted
Experts believe the dispute could jeopardise planned US-China discussions on AI safety scheduled for September.
According to analysts, additional sanctions or restrictions on Chinese AI developers could prompt retaliatory measures from Beijing, making diplomatic engagement more difficult.
Paul Triolo, a partner at DGA-Albright Stonebridge Group, said the scale of any US punitive action would determine whether broader diplomatic meetings, including a proposed summit between Presidents Donald Trump and Xi Jinping, remain on track.
China has also reportedly been considering restrictions on overseas access to some of its AI models as part of its response to US technology measures.
The disagreement builds on several years of tightening US export controls intended to limit China’s access to advanced semiconductor technology, an issue that continues to dominate economic and trade discussions between the world’s two largest economies.
Advanced AI models present new security challenges
Researchers call for international cooperation
Despite increasing political tensions, AI specialists argue that collaboration has become more important as frontier AI models continue to improve.
Attention has recently focused on a cybersecurity incident involving AI platform Hugging Face, which reportedly used Chinese company Z.ai’s GLM-5.2 model during efforts to contain a cyberattack involving an OpenAI testing agent. The incident has renewed debate over how advanced AI systems should be evaluated before deployment.
Researchers are also monitoring the emergence of recursive self-improvement, a capability allowing AI systems to enhance their own performance with reduced human intervention. Although still an evolving field, experts believe such developments reinforce the need for internationally recognised safety standards.
Open-weight AI models increase regulatory challenges
Particular concern surrounds open-weight AI models, which allow developers to download, modify and redistribute systems with relatively few technical restrictions.
Unlike closed commercial models, open-weight systems can be adapted independently after release, making it considerably harder for governments to enforce software-based safeguards or export controls.
Speaking remotely at a major AI conference in China, AI researcher Yoshua Bengio argued that developers should carefully assess models before publication and only release those falling below agreed risk thresholds.
He said stronger international evaluation processes could help reduce the dangers associated with increasingly powerful AI technologies.
Different priorities shape AI development in China and the United States
Resource limitations influence Chinese developers
Industry observers note that many Chinese AI laboratories operate with fewer computing resources than leading American companies such as OpenAI and Anthropic.
As a result, Chinese developers often prioritise improving model performance over extensive safety training, although government regulations require AI systems to undergo security and content reviews before public release.
However, existing Chinese rules primarily apply before deployment and do not fully address modifications made after models become publicly available.
Kristy Loke, a research fellow with the Machine Learning Alignment Theory Scholars (MATS) programme specialising in Chinese AI governance, suggested that China could eventually tighten policies governing open-weight frontier models, although significant regulatory changes would require substantial political support.
Debate continues over regulation of frontier AI
Some AI researchers argue that both China and the United States should introduce stronger mandatory testing before releasing advanced AI systems, alongside independent third-party safety evaluations.
In the United States, companies may voluntarily submit their models to the federally supported Center for AI Standards and Innovation (CAISI) for assessment. However, some lawmakers have called for these evaluations to become compulsory.
Loke said the most effective long-term approach would involve both countries agreeing on shared standards for pre-release testing while establishing clear limits for the deployment of the most advanced open AI models.
Washington remains divided over Chinese AI competition
American policymakers and technology leaders also remain divided over how to respond to the rapid growth of Chinese AI models, which now account for a substantial share of usage by US companies on the OpenRouter platform.
OpenAI and Anthropic have urged Washington to address competition from lower-cost Chinese models, arguing they pose commercial and strategic challenges.
Conversely, White House AI adviser David Sacks has argued that excessive regulation could weaken American competitiveness by limiting open-source innovation rather than strengthening it.
Conclusion
As artificial intelligence becomes increasingly powerful, the balance between technological competition and international cooperation is becoming more difficult to maintain. While the United States and China continue to compete for leadership in AI development, researchers on both sides argue that meaningful collaboration on safety standards may be essential to reducing the risks posed by the next generation of advanced AI systems.
Julian Barnes is an acclaimed British novelist, essayist, and short-story writer renowned for his elegant prose and intellectual depth. His work often explores themes of memory, history, love, and the complexities of human relationships. Widely regarded as one of the leading voices in contemporary British literature, Barnes has earned international recognition for his thoughtful and innovative storytelling.