The competition between the United States and China to build artificial general intelligence (AGI) is reshaping geopolitics in ways that could ultimately undermine both nations' security, according to defense analysts and AI researchers. Rather than a winner-take-all race, experts warn that the pursuit of AGI supremacy without international cooperation risks catastrophic outcomes, from autonomous weapons systems spiraling out of control to AI systems that neither democracy nor autocracy can safely govern.

Why Is the AI Race Being Compared to Nuclear Competition?

The stakes in the current AI rivalry echo Cold War anxieties about technological dominance. Vladimir Putin's assertion that "the one who becomes the leader in this sphere will be the ruler of the world" captures the existential framing that drives both superpowers forward. This perception of AI as a zero-sum competition has already triggered massive investments: the United States alone spent $18 billion on automated weapons systems between 2016 and 2020, a four-year window that illustrates how quickly the arms race is accelerating.

Science fiction has long warned about these dangers. Works like "Terminator" and "Alien" predicted not just the rise of robotics, but also the concentration of power in the hands of a "techno-elite" that could subvert representative government. These dystopian visions are no longer purely speculative. The integration of AI into China's Social Credit System and surveillance apparatus demonstrates how AGI could amplify either democratic values or authoritarian control on a global scale, depending on who develops it first.

What Happens If the Competition Continues Unchecked?

The danger lies not just in who wins, but in how the race itself transforms both competitors. Researchers Henry Farrell, Abraham Newman, and Jeremy Wallace observe that "AI will not transform the rivalry between powers so much as it will transform the rivals themselves". This transformation is already visible: AI is being weaponized for election interference, with 215 distinct incidents of deepfakes and misinformation campaigns documented globally during the 2024 election cycles in the United States, Taiwan, Indonesia, India, Pakistan, and Slovakia.

Henry Farrell

Both democracies and autocracies face a shared vulnerability that neither system has fully grappled with. Autocracies struggle to get honest feedback from their populations, which means AI trained on biased or false inputs will reinforce those distortions, creating a self-fulfilling prophecy that eventually destabilizes the regime. Democracies, meanwhile, are vulnerable to misinformed public feedback that AI systems then amplify and act upon. The result is that autonomous AI governance, regardless of the political system deploying it, poses an "unconscionable" risk to long-term stability.

How Can Nations Reduce the Risks of AI Competition?

Experts propose a fundamental shift in how the world approaches AGI development. Rather than treating it as a race, they argue for a cooperative framework that prioritizes safety over speed. Michael O'Hanlon, a defense analyst, contends that "safety must not only be an afterthought or viewed entirely in zero-sum terms," and that "all nations have an interest in preventing abuse of this century's transformative technology". He warns that "unrestrained competition could exacerbate the dangers".

The window for implementing such safeguards is narrowing rapidly. Nick Bostrom, a leading AI safety researcher, argues that humanity is in a "quickly shrinking window" in which AGI can be regulated before it reaches superintelligence, a point at which human control becomes theoretically impossible. Yoshua Bengio, one of the founders of modern deep learning, has warned that a "superintelligent yet badly regulated AI might even try to hoodwink a rogue actor" into producing and deploying it in dangerous ways.

• Establish International Oversight: Create binding agreements between the US and China that prioritize AGI safety over competitive advantage, similar to nuclear non-proliferation treaties.
• Slow Development Timelines: Implement coordinated pauses in AGI research to allow safety measures to catch up with capability advances, reducing the risk of uncontrolled systems.
• Transparent Governance Frameworks: Require that AI systems be built with explainable decision-making processes and human oversight mechanisms, regardless of which nation develops them.
• Global Safety Standards: Establish international standards for testing and deploying AGI that apply equally to all nations, preventing a race to the bottom on safety requirements.

The core tension is that slowing progress in the United States risks allowing China to eclipse American AI capabilities in a short timeframe, triggering what analysts call a "downward spiral of a security dilemma". Yet accelerating without safeguards risks creating an AGI system that neither the US nor China can control, let alone govern responsibly. Warnings from figures as diverse as Elon Musk, Bill Gates, Henry Kissinger, and Stephen Hawking that "we will neither be able to understand it nor control it" remain largely unheeded in the rush for dominance.

The paradox is that winning the AI race may ultimately mean losing the ability to govern the technology safely. Both superpowers face the same fundamental challenge: building AGI systems that reflect their values while remaining controllable and transparent. Without a paradigm shift toward cooperation, the race itself may become the greatest existential risk of all.