For years, experts have predicted that artificial intelligence would revolutionize world politics. History may record 2026 as the moment the revolution arrived. From the start of this year, continuous developments have underscored the possibilities and perils of the AI era, from the use of AI in U.S. military interventions to the emergence of AI cyberweapons that can significantly outperform human operators. “Loss of control” episodes—cases in which AI models escape human control and supervision—have become more common. U.S. officials, including President Donald Trump, tout the American lead at the frontier of AI development, even as they worry about what dangers lurk over the horizon.

Some experts believe that humanity is within striking distance of artificial general intelligence: technology capable of performing a wide array of tasks as well as or better than humans. There is growing speculation that artificial superintelligence, or ASI—a slippery concept, loosely defined as AI models that vastly exceed human capabilities across all tasks and domains—could arrive within a few years. Today’s top AI models, by contrast, surpass human capabilities only in narrow areas such as mathematics and data processing and lag human capabilities in areas such as common sense and deep reasoning. The emergence of ASI might bring breakneck economic and scientific progress or unlock game-changing military advances that upend the global balance of power. But it could also fundamentally shift the relationship between machines and humanity, perhaps, in the most dire scenarios, allowing near-omnipotent systems to eradicate their human creators.

Much of this is hypothetical. It’s still uncertain whether ASI is possible, when it might be developed, or whether it could be kept aligned with the flourishing of humanity. These uncertainties dominate discussions of any policy response. But because the implications of ASI would be seismic, it is important to assess how the United States might approach the technology and how different ASI strategies would affect Washington’s engagement with the world.

One option is to pursue supremacy by sprinting toward ASI while holding rivals back. This strategy rests on the belief that the greatest danger to the United States is not uncontrollable ASI technology but a rival controlling that technology. Another option is shared development, in which the United States seeks to responsibly unlock the benefits of ASI for humanity—and tamp down the dangerous dynamics of a race to build it—by developing it within a safety-conscious consortium of states. A third option is suppression: the United States would forgo development of ASI and stop anyone else from building it on the basis that superintelligence is just too dangerous to be unleashed on the world.

Each of these options prioritizes different aspects of the ASI challenge—the quest for geopolitical advantage, the need to safely develop a revolutionary technology, the importance of minimizing existential dangers. Yet each also contains serious risks or flaws and requires making potentially irrevocable bets on a currently unknowable future. If the underlying premise driving a particular strategy proves wrong, that strategy could fail in catastrophic ways. It would thus be a mistake for U.S. leaders to fully embrace any of these options right now. The best near-term option is prudent hedging, to position the United States to take advantage of a variety of possible futures while informing its ability, eventually, to make a clearer choice.

The urgency of the ASI debate flows from surging achievements and expectations. Top AI models have become increasingly adept at solving complex problems across a growing number of domains. Some show superhuman capabilities in areas such as cyberwarfare. Leading experts speculate about a coming “intelligence explosion”—the point at which AI becomes capable of rapidly and continually improving itself. The widely read AI 2027 scenario—a 2025 forecast, developed by the nonprofit AI Futures Project, that depicts step by step how a fast takeoff could unfold—hypothesizes that ASI could emerge in this decade. Likewise, Dario Amodei, co-founder of Anthropic, has written that “humanity is about to be handed almost unimaginable power.”

If ASI does materialize, the impacts could be epochal. ASI optimists predict that it could conquer climate change by unlocking new carbon-capture technologies or cure cancer by resolving the mysteries of human biology. ASI-enabled breakthroughs in propulsion could allow deep-space travel and planetary colonization. Yet this tremendous economic and scientific promise exists alongside potentially epic disruptions. ASI might empower bad actors by enabling devastating cyberattacks and bioweapons; usher in the automation of military decision-making, causing wars to become faster and less controllable; or unlock wonder-weapons, such as sensors that render the oceans effectively transparent. It could transform the balance of power by lending a massive military advantage to countries that achieve ASI first. There’s no guarantee that these advantages would fall to democratic societies. Amodei warns that ASI could cause a “totalitarian nightmare” by enabling all-encompassing surveillance and automating repression.

The gravest risk is that superintelligence could enslave or exterminate humanity. ASI might help the human race destroy itself by helping nihilists unleash novel pandemics or cause automated “flash wars” that result in nuclear catastrophe. Or perhaps a technology smarter than its creators will ultimately come to view them as threats to be eliminated or resources to be exploited. Some AI executives put this “probability of doom” as high as 25 or even 50 percent.

None of this forecasting is certain. Some experts doubt that ASI is technically possible or argue that material impediments, such as limitations on computing power (or “compute”) and electricity, could forestall its arrival. Others question whether ASI will produce lasting, first-mover advantages, believing that its economic and military benefits, like those of most prior technologies, will accrue in gradual and uneven ways. Leading observers also disagree on the possibility of........

© Foreign Affairs