America is in the wrong AI race with China

In the U.S., the “AI race” is understood as a distinctly anti-China battle, framing model performance as the deciding factor. U.S. policymakers have, across multiple administrations, shaped the country’s entire regulatory stance to be distinctly light-touch, underpinned by the belief that regulation could lose us this global battle for artificial intelligence dominance. What if they’re wrong? Recently, OpenAI CEO Sam Altman admitted the AI revolution was going slower than expected, blaming “human habits and institutional inertia.” The data tells a different story: Americans increasingly distrust the technology even as AI companies race to place it at the center of everyday life. A majority of Americans are now more concerned than excited about AI, while trust in the government to regulate AI effectively remains weak. The problem is not confined to abstract anxiety about the future. China has recognized that governing AI means governing its deployment, not merely the capabilities of its models. AI is beginning to reshape electoral politics, becoming the hot-button issue for the U.S. midterm elections. Conservative voters, angry about the water, electricity, and land demands of AI infrastructure such as data centers, say they are preparing to “break rank.” The technology industry has discovered a basic political truth: People will not support an AI future if they believe its costs are being imposed on them while its benefits accrue elsewhere. In stark contrast, what little conversation is entertained in Washington focuses on narratives of “uncontrollable rogue agents” and loss of control. Consumer protection is too often treated as the mundane cousin of AI safety. Meanwhile, lawsuits accumulate on harmful manipulation, AI toys marketed to children surface explicit content, commonly available AI models easily allow nonconsensual deepfakes, and AI introduced in online maps amplifies misinformation. Effective consumer protections give users agency: the ability to know when they are interacting with an AI system, understand when AI has generated or altered information, challenge consequential automated decisions, and seek redress when things go wrong. Absent these protections, younger “digital native” Americans are clear: They will not willingly purchase or use these products. A formidable competitor There is an uncomfortable geopolitical irony here. China seems to be engaging in a different AI race altogether. The Chinese government has already built a more extensive regulatory environment for governing how the technology interacts with users. While its approach is hardly a model for liberal democracies — China’s rules are embedded in a system of extensive state control and censorship — it would be a mistake to ignore the competitive lesson. China has recognized that governing AI means governing its deployment, not merely the capabilities of its models, treating AI as bureaucratic technology. For example, rules on labeling AI-generated content require transparency across the AI content ecosystem and seek to make synthetic material easily identifiable to users. Similarly, laws protect vulnerable populations, such as children or the elderly, from anthropomorphic AI manipulation and harm, and consumers from algorithmic price discrimination. Cheaper or free open-source models are preferred by companies becoming more conscious of their AI consumption costs. China is pursuing these rules while remaining a formidable competitor to the U.S. in advanced AI development and adoption. There is also a greater sense of optimism: 87% of Chinese citizens trust AI compared to 23% in the U.S. It seems that performance at all costs isn’t motivating adoption by American companies, either. The OpenAI Sora deepfake debacle cost the company a $1-billion licensing deal with Disney. Airbnb CEO Brian Chesky admitted his company uses Alibaba’s Qwen model because “it’s very good. It’s also fast and cheap.” And Anthropic’s latest Fable model has had the slowest adoption to date, as cheaper or free open-source models are preferred by companies becoming more conscious of their AI consumption costs. Defining who wins the AI race is less clear than many in Washington or Silicon Valley suggest. This is particularly interesting as the rest of the world commits heavy investment in their sovereign AI stack. Most dismiss Europe as a serious contender, arguing the region can never outperform the U.S. or China in model performance. However, the European Union can demonstrate that its long-standing tradition of consumer protection may be the key to sustainable innovation. And, as countries from Singapore to India and Kenya meet to discuss and define AI sovereignty, they should resist equating it with compute: Lasting autonomy rests on trust, not teraflops. It seems neither consumers nor companies are making choices about AI based on model performance alone. Determining the winner of the global AI race will be less about spectacular demonstrations and more about intimate deployment. Companies will use AI to recommend financial products, advise patients, assess job applicants, help children learn, and make decisions on behalf of consumers. This will only happen if the public feels model developers and governments have adequately addressed their concerns. All of this will matter as the 81st session of the United Nations General Assembly gets underway in New York City this week, against a backdrop of technological friction and mistrust. With Chinese President Xi Jinping scheduled to hold talks with Donald Trump, but opting to skip the general debate, and a difficult relationship between the U.N. and the Trump administration, other nations may have an opportunity to forge a better path forward.