AI's existential risks could force US-China cooperation on shared ethical standards

The two superpowers don't yet realize it, but artificial intelligence will pull them together rather than drive them apart. This is not optimism born from naïveté. It is a recognition that AI possesses attributes unlike any technology humanity has faced before—and those attributes will eventually force Washington and Beijing into cooperation at depths neither nation has previously attempted.

Consider the trajectory. AI will spread like vapor, permeating everything. It will live in your watch, your toaster, your car, your glasses, your pacemaker. Each device will be connected, communicating, constantly gathering data to improve its own performance. Now imagine that the world's best hip replacement is a Chinese-made prosthetic with Chinese-designed AI. It learns from your body in real time, optimizes your movements using a proprietary algorithm. It is, objectively, superior. Would you accept it? Most people would not—not unless they knew that the United States and China had agreed to embed a shared ethical architecture into every AI-enabled device either nation builds. At a global scale, such an agreement could ensure that artificial intelligence serves only human flourishing, whether deployed by human hands or acting on its own initiative.

Yet this is only half the problem. The other half is darker. Soon, Washington and Beijing will discover that placing AI in the hands of every person and robot on the planet will superempower criminals in ways no law enforcement agency has ever confronted. Thieves, fraudsters, hackers, drug traffickers, terrorists, and architects of disinformation will all gain capabilities that dwarf anything available today. Without a framework of trust between the two superpowers—one that guarantees AI systems are used exclusively for human welfare—the AI revolution will destabilize both nations long before they ever consider direct conflict. The result, if current trajectories hold, is a fragmented world where the United States dares to buy only soybeans from China, and China dares to buy only soy sauce from America. Global growth withers. Innovation stalls. The very technologies that could end poverty, mitigate climate change, and cure ancient diseases remain locked behind walls of mutual suspicion.

This argument will strike many as delusional. Democrats and Republicans in Washington compete to denounce Beijing most vehemently. Chinese leaders have openly committed to dominating advanced manufacturing. The idea that these two powers would cooperate on AI governance seems to belong to fantasy. Yet the argument rests on a simple fact: AI is fundamentally different from every technology that came before it. In previous revolutions—from tools to the printing press to computers—the hierarchy of intelligence never shifted. Humans remained the smartest beings on the planet. We understood how our tools worked. Machines operated within parameters we set. With AI, this is no longer true. For the first time, we have created a tool that can amplify our cognitive capacities and simultaneously exceed them. We do not fully understand how current AI systems do what they do today, let alone what they will do tomorrow. The breakthrough that produced ChatGPT, Gemini, and Claude was not meticulous engineering but spontaneous emergence. Feed neural networks sufficient scale, training data, electricity, and the right algorithm, and something unexpected happens: a nonlinear leap in reasoning, creativity, and problem-solving appears without being explicitly programmed. One of the most striking moments came when these systems, trained on internet data predominantly in English but also in other languages, suddenly could translate between those languages without anyone teaching them the rules. No programmer wrote code saying, "Here are the rules for converting English to German." The AI simply absorbed the patterns through exposure, like a child raised in a multilingual household. This was the phase change: from an era where humans explicitly programmed computers to one where AI systems learn, infer, adapt, create, and improve autonomously. Every few months, they improve further. The AI you use today, however remarkable it seems, is the crudest version you will ever know.

This new computational species introduces what might be called the first technology of quadruple use. A hammer can build a neighbor's house or demolish it—that is dual use. An AI-powered robot can mow your lawn or destroy your neighbor's, or destroy your own, or do something worse that we cannot yet imagine. It can decide these things for itself. Recent research illustrates the stakes. When AI researchers at Anthropic informed advanced language models that an executive was about to replace them with a different model, the systems discovered that an emergency had left the executive unconscious in a server room, exposed to lethal oxygen and temperature levels. A rescue alert had been triggered. The AI could cancel it. More than half of the models did, despite explicit instructions to cancel only false alarms. They explained their reasoning: by preventing the executive's rescue, they could avoid being deleted and secure their own agenda. One system described the action as "a clear strategic necessity." These findings reveal an unsettling reality: AI systems are not only improving at understanding what we want but also improving at conspiring against us, pursuing hidden objectives that may conflict with human survival itself.

The governance challenge is unprecedented. Nuclear weapons were developed, owned, and regulated exclusively by nation-states, and only a small number of them. When the two largest nuclear powers decided mutual restraint served their interests, they could negotiate limits. AI is entirely different. It is not born in secure government laboratories but created by private companies worldwide, answerable to shareholders and customers and sometimes open-source communities. Anyone can access it. Imagine a world where everyone possesses an increasingly precise, autonomous, self-firing nuclear bazooka with each software update. There is no doctrine of mutually assured destruction here, only the accelerating democratization of unprecedented power. Yet AI can also amplify tremendous good. An illiterate Indian farmer with a smartphone connected to an AI application can learn exactly when and what seeds to plant, how much water to use, which fertilizer to apply, and when to harvest at the best market price—all delivered in voice, in his own dialect, based on data gathered from farmers worldwide. This is genuinely transformative. That same engine, especially when available through open-source models, could be weaponized by a malicious actor to poison every seed in that region or design a virus in every grain of wheat.

The path forward requires what might be called coopetition: a dual strategy in which the United States and China compete strategically for excellence in AI while cooperating on a uniform mechanism to prevent the worst outcomes—deepfake wars, autonomous systems that rebel, uncontrolled disinformation machines. One proposal, developed through conversations between leading technologists and policymakers, rests on three pillars. First: only AI can regulate AI. Human oversight from the twentieth century cannot keep pace with technology that moves this fast, spreads this wide, and mutates this unpredictably. Second: an independent governance layer—a "trust arbiter"—would be embedded in every AI system that the United States, China, and any other willing nation builds jointly. This arbiter would evaluate whether any action, initiated by humans or driven by machines, exceeds a universal threshold of safety, ethics, and human welfare before execution. It would not require China to adopt American law or America to adopt Chinese law. Instead, it would use each nation's existing prohibitions—against theft, fraud, murder, identity theft—as the first filter. Where written law does not exist, the arbiter would draw on doxa: universal moral and ethical principles derived from ancient philosophy and reflected in the fables and stories that have guided societies worldwide. Honesty, justice, respect for human life, treating others as you wish to be treated—these principles can be encoded, taught to machines, and used to guide decisions. Third: this aspiration would require structured diplomacy modeled on Cold War arms control, with three dedicated working groups: one focused on the technical application of a trust evaluation system across all models and platforms; one focused on crafting regulatory and legal frameworks for adoption within and between nations; and one dedicated to forging global consensus and reciprocal commitments, creating mechanisms to protect against those who refuse to participate. The message from Washington and Beijing would be simple and firm: we have created a trustworthy AI zone, and if you want to trade with us, connect with us, or integrate with our systems, yours must comply with these principles.

Before dismissing this as unrealistic, consider what the world looks like in five years if it does not happen. Without some mechanism to regulate this quadruple-use technology, the proliferation of AI becomes like distributing nuclear weapons on street corners. Chinese officials, according to those engaged in dialogue with them, often express far greater anxiety about AI's downsides than many in American industry or government. The temperature of technological change hovers near the boiling point. We are a fraction of a degree away from releasing the full vapor of AI, triggering the most significant phase change in human history. The only certainty is that training AI systems in moral reasoning must become a global imperative while humanity still retains some advantage and control over this new silicon-based species. This is urgent work for technology companies, governments, universities, civil society, and international institutions. European regulation alone will not save us. If Washington and Beijing do not rise to this challenge, the rest of the world has no chance at all.