The intensifying technological rivalry between the United States and China is threatening efforts to establish common safeguards for artificial intelligence, as both countries prepare for planned AI talks later this year while simultaneously accusing each other of unfair practices and security risks.
Tensions escalated after U.S. officials accused Chinese AI startup Moonshot of using outputs from Anthropic’s advanced Fable 5 model to train its own Kimi K3 system through a technique known as AI distillation. Treasury Secretary Scott Bessent warned that sanctions against Chinese AI developers remain under consideration, while the U.S. Commerce Department is investigating whether Chinese firms have illegally accessed advanced American semiconductors restricted under export controls.
Moonshot did not respond to requests for comment.
The dispute highlights a growing contradiction in the AI race. Both Washington and Beijing increasingly see frontier AI systems as strategic national assets with military, economic, and cybersecurity implications. At the same time, experts argue that cooperation between the world’s two largest AI powers has become more necessary than ever as advanced models gain capabilities that could be exploited for cyberattacks, critical infrastructure disruption, or autonomous decision making.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
Planned AI Dialogue Faces Fresh Uncertainty
The accusations against Chinese developers, combined with reports that Washington is considering tighter restrictions on Chinese open weight AI models, could undermine plans for a bilateral AI dialogue expected in September.
Analysts warn that any new sanctions could trigger reciprocal measures from Beijing and spill over into broader diplomatic relations, including a planned meeting between U.S. President Donald Trump and Chinese President Xi Jinping.
Paul Triolo, a partner at DGA Albright Stonebridge Group, said extensive punitive measures against Chinese AI firms could derail both the planned AI discussions and the broader diplomatic agenda between the two countries.
Meanwhile, China is reportedly considering restrictions on overseas access to some of its most advanced AI models as policymakers weigh responses to possible U.S. actions.
Growing Safety Risks from Advanced AI
The debate comes as AI capabilities continue advancing rapidly.
Last week, AI platform Hugging Face reportedly relied on Chinese developed model GLM 5.2 from startup Z.ai during testing involving a rogue OpenAI agent, arguing that some American closed models contained safety restrictions that limited their effectiveness in certain defensive cybersecurity scenarios.
Researchers increasingly warn that frontier AI systems may eventually reach recursive self improvement, allowing them to autonomously improve their own capabilities. Such advances would significantly increase the importance of internationally accepted safety standards.
Experts also warn about open weight AI models, which allow users to freely download, modify, and redistribute the underlying model. While these systems encourage innovation and broader research, they also make it easier for malicious actors to remove safety protections.
Speaking via video link at China’s leading AI conference last week, AI pioneer Yoshua Bengio warned that once highly capable open weight models are widely distributed, restricting their misuse becomes extremely difficult.
He argued that governments should develop rigorous evaluation systems to determine which models can safely be released while restricting those that exceed acceptable risk thresholds.
Different Regulatory Approaches
China currently requires domestic AI models to undergo government content and safety reviews before public release, although existing regulations generally focus on initial deployment rather than later modifications.
Industry observers note that many Chinese AI companies operate with fewer computing resources than American leaders such as OpenAI and Anthropic. As a result, they often prioritize improving model performance over extensive safety testing.
In the United States, frontier AI developers can voluntarily submit models for evaluation by the federally supported Center for AI Standards and Innovation, although some lawmakers have proposed making independent safety reviews mandatory before deployment.
Researchers argue that stronger international cooperation could produce shared testing standards, independent evaluations, and agreed red lines governing the release of the world’s most advanced AI systems.
Washington Divided Over Chinese AI
The rapid growth of Chinese AI models has also sparked debate within the United States over how aggressively they should be restricted.
Chinese models reportedly account for roughly 60 percent of token usage by U.S. companies using the OpenRouter platform, reflecting their growing competitiveness and lower operating costs.
OpenAI and Anthropic have urged policymakers to address the competitive challenge posed by Chinese developers, arguing that unrestricted adoption could weaken America’s technological leadership.
Others disagree.
White House AI adviser David Sacks argued that American AI companies should compete through innovation rather than government intervention. He suggested some firms are seeking regulatory barriers against lower cost competitors instead of relying on market competition.
Why It Matters
The dispute reflects the increasingly complex balance between technological competition and global security. Both the United States and China recognize that advanced AI could become one of the defining strategic technologies of the century, yet meaningful safeguards may require cooperation between rivals.
Whether planned AI talks proceed despite growing political tensions could influence not only future regulation of frontier AI systems but also broader U.S. China relations at a time when competition extends from semiconductors and trade to military technology and cybersecurity.
Analysis
The current dispute illustrates a fundamental dilemma in global AI governance. Strategic rivalry encourages both Washington and Beijing to accelerate AI development while limiting the transfer of technology to competitors. However, the same race increases the risks associated with increasingly powerful AI systems. If political tensions derail bilateral safety discussions, both countries may pursue separate regulatory frameworks with limited transparency, making it more difficult to establish shared rules for evaluating frontier AI models. While competition is likely to remain the dominant feature of U.S. China relations, maintaining technical dialogue on AI safety may become one of the few areas where cooperation serves the strategic interests of both sides.
With information from Reuters.
