Why Are the US and China Struggling to Agree on How AI Should Be Controlled?

The United States and China both recognise that advanced artificial intelligence carries serious risks, yet they have made little progress toward shared rules or mutual restraints. The difficulty is not a simple lack of technical understanding. It stems from clashing strategic priorities, divergent views of what constitutes the main danger, deep mutual distrust, and incompatible ideas about who should set the rules.

Strategic Competition Overrides Shared Risk

Both governments treat leadership in AI as a core element of national power. American officials repeatedly frame the contest as one the United States must win in order to protect its economic and military edge. Chinese officials view rapid development as essential to closing the gap and reject any framework that appears designed to freeze the existing hierarchy. When one side proposes slowing the frontier or limiting certain capabilities, the other often interprets the suggestion as an attempt to lock in advantage rather than a genuine safety measure.

This zero-sum lens makes cooperation fragile. Even when researchers and some officials on both sides acknowledge overlapping concerns about loss of control, accidents, or misuse, the political incentive remains to keep pushing capability forward. A unilateral pause is seen as self-defeating; a mutual pause requires a level of verified restraint that neither side currently trusts the other to deliver.

Different Definitions of the Problem

The two countries emphasise different categories of risk. In the United States, public and expert debate frequently centres on long-term or existential scenarios: highly autonomous systems that could escape meaningful human control, catastrophic misuse, or rapid capability jumps that outpace governance. Safety discussions often focus on frontier model evaluation, compute thresholds, and responsible scaling.

Chinese policy and commentary tend to prioritise nearer-term and socially grounded risks. These include the spread of politically sensitive or false content, algorithmic manipulation, cyber threats, and the preservation of state authority over data and information flows. Beijing has already imposed domestic rules requiring labelling of AI-generated content and tighter oversight of recommendation algorithms. From this perspective, external pressure to slow overall development can look less like risk management and more like an effort to constrain a competitor.

Because the two sides rank the dangers differently, they struggle to agree on which behaviours should be restricted first or how compliance would be measured.

Trust Deficit and Accusations of Bad Faith

Years of technology controls, intellectual-property disputes, and reciprocal accusations have eroded confidence. The United States has restricted exports of advanced AI chips and related tools, arguing that these measures protect national security. China describes the same controls as an attempt to build an “AI iron curtain” and maintain monopoly power. American officials have accused Chinese entities of improperly extracting capabilities from leading Western models through distillation and other techniques. Chinese responses frame such charges as pretexts for containment.

Past experiences in other domains reinforce scepticism. Each side can point to instances in which the other is perceived to have under-delivered on earlier understandings. The result is a credibility problem: proposals for verification, hotlines, or red lines are filtered through the assumption that the counterpart will exploit any restraint.

Divergent Regulatory Philosophies

Domestic approaches to AI also diverge. The current U.S. administration has resisted comprehensive new federal regulation, arguing that heavy rules could handicap American innovation precisely when speed matters most. Industry voices sometimes call for clearer guardrails, yet political leadership has emphasised leadership through capability rather than through restriction.

China already exercises stronger state direction over algorithms, training data, and content outputs. At the same time, it rejects externally imposed limits on the pace of model development and insists that any global framework must respect sovereignty and give developing countries a meaningful voice. These contrasting starting points make it hard to design rules that both governments would accept as binding on their own companies and laboratories.

Practical Obstacles to Verification

Even if political will existed, technical and practical barriers remain. Training runs can be distributed, models can be refined through distillation or synthetic data, and dual-use applications are difficult to monitor across borders. Open-weight models, which China has promoted as a route to wider adoption, complicate control because capabilities can spread once weights are released. Closed models raise different transparency problems. Without agreed metrics, inspection rights, or trusted third-party evaluation, any formal commitment would be hard to enforce.

Narrow Areas of Possible Overlap

Despite the obstacles, limited cooperation is not impossible. Both sides have an interest in preventing accidents that could be misread as deliberate attacks, especially around nuclear command systems or large-scale autonomous cyber operations. Technical dialogues on incident communication, basic safety testing standards, or shared understanding of certain catastrophic failure modes have been discussed by experts. Progress on these narrower topics would still require political space that current rivalry has so far limited.

The United States and China struggle to agree on AI control because each sees the technology as decisive for future power, ranks the primary risks differently, distrusts the other’s motives and compliance, and operates from distinct regulatory traditions. Until the strategic competition softens or a crisis forces narrower, verifiable understandings, comprehensive joint governance is likely to remain elusive.

What aspect of the US-China AI disagreement do you find most consequential for global safety or innovation? Share your view.

Frequently Asked Questions

Why is AI treated as a national security issue by both countries?
Both governments believe advanced AI will shape economic productivity, military capability, and geopolitical influence, making leadership a strategic priority.

Do the US and China worry about the same AI risks?
Not entirely. US discussions often emphasise long-term loss-of-control and frontier safety; Chinese policy focuses more on content control, social stability, cyber risks, and sovereignty.

What role do export controls play?
US restrictions on advanced chips aim to limit China’s access to cutting-edge computing. China views them as containment measures that justify faster indigenous development.

Is any cooperation occurring?
Expert-level and some official dialogues have explored safety topics, but political distrust and competing priorities have kept formal agreements limited.

Why is verification so difficult?
Model training can be opaque, capabilities can transfer through distillation or open releases, and dual-use applications are hard to monitor across jurisdictions.

Could a crisis change the dynamic?
A serious incident involving uncontrolled AI systems might create pressure for narrow, practical understandings, but it would not automatically resolve the deeper strategic rivalry.

Reference links:
https://www.independent.co.uk/news/china-donald-trump-xi-
jinping-beijing-washington-b3051628.html
https://www.reuters.com/legal/litigation/ai-rivalry-hangs-over-trump-xi-talks-2026-09-16/
https://www.ft.com/content/83023f2f-0c12-4239-bf27-99d8e378ec5d
https://asiatimes.com/2026/09/why-a-us-china-ai-regulation-deal-is-nowhere-in-sight/
https://www.reuters.com/legal/litigation/ai-grows-more-
powerful-us-china-feud-threatens-safety-efforts-2026-07-24/
https://www.cbsnews.com/news/us-china-ai-development-race-mutual-distrust-dangers/