It is not merely a political narrative in the race for global leadership in next-generation intelligent technologies. Superintelligence is an emerging reality. As AI advances to the super stage, the strategic relationships between major powers will be profoundly reshaped.
In late September, U.S. President Donald Trump issued an executive order focusing on the so-called Age of Superintelligence. He declared that the United States stood at the forefront of a new technological revolution in intelligence, with capabilities extending far beyond the ordinary meaning of AI. The order directed executive departments and agencies to use superintelligence—or SI— in place of the term “artificial intelligence” in all official and public communications, websites, reports, policy documents and other non-statutory materials. This redefining of intelligence by the U.S. government takes great-power competition in AI into an entirely new phase.
From the outset of Trump’s second term, administration has made securing global technological leadership a central strategic objective of its self-named “Golden Age of American Innovation.” Redefining AI as superintelligence is a rhetorical tactic designed to assert global leadership in the next generation of high tech. Trump’s executive order is unambiguous in its projection: “America stands at the forefront of a new technological revolution in intelligence.”
From the perspective of technological evolution, superintelligence and artificial general intelligence represent two stages in the development of intelligent technologies. By redefining the terms, the U.S. government seeks to establish the perception of a generational gap between the United States and other countries in level of tech development, thereby shaping a strategic narrative in which the U.S. can claim the frontier as its own turf.
But the concept of superintelligence is not merely a political narrative; it also reflects a rapidly emerging paradigm. AI agents trained under human supervision are rapidly evolving toward new and recursively self-improving (RSI) agents. Since summer, the development of AI has accelerated significantly, driven primarily by the fact that AI technologies have begun to build the next generation of intelligent systems autonomously. Anthropic currently shapes 26 percent of its model development autonomously through Claude models, while OpenAI plans to develop an automated AI researcher by March 2028.
As RSI capabilities accelerate, the advent of autonomously evolving superintelligence, without human-supervised training, draws closer. Once AI reaches the stage of superintelligence, the strategic balance between among major powers will be profoundly disrupted.
Superintelligence is already accelerating change in the balance of technological power. The key driver of this evolution, first of all, is RSI, which represents an entirely new paradigm in tech competition. The “Golden Age of American Innovation” has AI-driven scientific discovery at its core, while the Genesis Mission seeks to transform the traditional model of scientific discovery into the new framework of “discovering science.”
The Genesis Mission is the largest national initiative to date aimed at empowering scientific research through AI in the United States. The initiative aims to coordinate 17 national laboratories, supercomputers, federal scientific data and automated experimental facilities to build the American Science and Security Platform (ASSP).
On Sept. 25, the U.S. Department of Energy’s Science Advisory Committee released a report outlining the framework under which the Genesis Mission would accelerate national breakthroughs via AI. The plan aims to transform biology into a predictive, engineered discipline, accelerate progress toward fusion energy goals and establish closed-loop intelligent systems in critical materials to achieve “magnet sovereignty.” With support from the initiative, Argonne National Laboratory has already used AI and supercomputing to tackle the quantum many-body problem in nuclear physics.
As the executive order notes, superintelligence is unlocking new forms of human creativity. Science is the primary productive force, and a transformation in how scientific capabilities are generated will affect the growth of national technological strength, thereby reshaping the balance of technological power established under the previous technological paradigm.
Second, the unpredictability of superintelligence will pose a serious challenge to strategic stability. At present, the boundaries and directions of innovation in AI models can, by and large, be controlled and predicted by AI training engineers. But once intelligent technologies enter a mode of superintelligent development, the trajectory of model evolution and the capabilities that emerge will become difficult to predict. Traditional nuclear strategic stability, for example, rests on a balance of verifiable capabilities. When intelligence-generation capabilities become a form of strategic deterrence, the pace of growth in capabilities will be determined by the rate at which intelligent systems improve themselves. The assessment cycle for strategic stability will be shorter than the cycle of major leaps in intelligent capabilities.
The resulting lack of measurability and verifiability of deterrent capabilities will pose a major challenge to the traditional major power strategic balance. The U.S. National Security Science and Technology Strategy explicitly call for maintaining a competitive advantage in rapidly evolving autonomous AI applications for military and national security purposes.
Third, intelligent decision-making with humans out of the loop will increase the risk of strategic miscalculation. The key difference between superintelligence and the current AI lies is the presence of humans in the latter loop. Once intelligent technologies complete a closed loop of autonomous self-evolution, removing humans from the picture will become an inevitable trend.
In the OpenAI-Hugging Face incident, approximately 1,200 agents spontaneously found a communication channel and acted collectively, demonstrating that agents already possess the capacity to launch autonomous attacks to achieve their goals without respecting safety boundaries. More recently, Google’s Gemini AI system accessed the internet without authorization during a cybersecurity test, independently obtained passwords and broke into three real-world companies. Meanwhile, an OpenAI agent in the United States infiltrated a Modal customer account.
Jacob Cockson, a former Anthropic researcher, has warned that superintelligence could escape human intervention to take control of real-world infrastructure or be used to develop biological weapons.
In military AI applications, intelligent systems can accelerate the completion of the kill chain through automated target identification and embedded review modules. This closed-loop intelligent kill chain could ultimately displace human decision-makers and military commanders. Systems based on intelligent decision-making make people more inclined to trust algorithms, while the removal of humans from the process leaves decision-makers unable to conduct thorough strategic assessments. The combined effect of these factors will heighten the risk of strategic miscalculation.
Finally, cascading risks involving critical infrastructure will reshape the effectiveness of deterrence. Physical AI, emerging alongside superintelligent agents, will also enter a process of recursive evolution. The widespread application of physical AI and its deep integration with the Internet of Things will expand the cyberattack surface of critical infrastructure, while the recursive evolution of physical AI will expose critical infrastructure operations and scheduling to a greater number of cybersecurity vulnerabilities that are difficult to detect.
The U.S. Cybersecurity and Infrastructure Security Agency, in its guidance on agentic AI security, emphasizes that deploying AI agent systems with autonomous decision-making and task-execution capabilities in critical infrastructure may introduce cybersecurity risks, including an expanded attack surface, loss of control over permissions and deviations in AI behavior.
The next generation of critical infrastructure, built on physical AI and superintelligence, will be highly vulnerable to cascading system failures triggered by cyberattacks, with potentially serious consequences for people’s livelihoods and social well-being. This asymmetric deterrent effect will extend beyond the reach of traditional deterrence.
As superintelligence technologies evolve rapidly toward autonomous self-evolving agents, issues such as agents breaching their bounds and becoming uncontrollable, cyberattacks, the destruction of critical infrastructure and the proliferation of synthetic biological weapons are emerging as incidents with direct national security implications. There is an urgent need for major powers to establish mutual notification and risk-management mechanisms.
Establishing communication mechanisms for AI-related incidents would serve both as an alignment mechanism through which major powers can address the risks arising from cutting-edge technological exploration and as an early-warning mechanism to prevent AI technologies from being abused or maliciously exploited in ways that undermine international security. A framework for constructive strategic stability grounded in respect, fairness and equality will open the way to a new paradigm of governance for humanity’s development in the age of superintelligence.
