Incentives for U.S.-China Conflict, Competition, and Cooperation Across Artificial General Intelligence’s Five Hard National Security Problems

We think that it is important that the United States and its allies and partners proactively work with China to shape the terms of AI competition and cooperation. A future in which uncontested AI escalation or miscalculation could produce crises that no side can control is not in the United States’ national interest, nor is it in the PRC’s. Strategic, calculated engagement—balancing competition with measured, issue-focused cooperation—may reduce the risk of strategic surprise, WMD proliferation, and catastrophic conflict, while protecting U.S. interests and global stability. This, in turn, will allow the U.S. government to cultivate mechanisms and norms for discussion and cooperation around AGI-related issues.

U.S.-PRC Strategic Rivalry: Mutual Threat Perceptions and Competitive Dynamics
Policymakers in Washington view the PRC as the main national security challenge confronting the United States, and U.S. defense strategists are becoming more and more focused on how to counter the growing capabilities of China’s People’s Liberation Army (PLA) (Colby, 2021). In particular, there is growing concern that the PRC may seek to use force or coercion to annex Taiwan, and U.S. analysts are closely tracking the PLA’s development of military capabilities designed to exploit potential U.S. vulnerabilities in the event of a conflict. Moreover, some U.S. analysts judge that the PRC seeks to displace the United States in the Indo-Pacific, and ultimately to eclipse U.S. influence globally (Doshi, 2023). The United States has also become increasingly focused on economic and technological competition with the PRC, and Washington has sought to reduce dependence on PRC supply chains and to constrain PRC technological advances perceived as threatening to U.S. economic and security interests.

Likewise, PRC perceptions are based on deep suspicion of U.S. intentions, with analysts in the PRC judging that the United States views almost all aspects of its relationship with China through a security lens and is determined to undermine the Chinese Communist Party (CCP), slow China’s economic and science and technology advancement, and ultimately to prevent China from achieving its rightful place on the global stage. Xi Jinping, who serves concurrently as CCP General Secretary, President, and Central Military Commission Chairman, has stated publicly that the United States and its allies and partners are determined to ramp up their “containment, suppression, and encirclement” of the PRC (Bradsher, 2023).

China’s May 2025 national security white paper paints a similar picture, suggesting that the PRC must contend with external pressure and instability that it attributes mainly to the United States, and specifically highlighting the United States strengthening its alliances, deploying intermediate-range missiles, and adjusting its military posture in the region (State Council Information Office of the People’s Republic of China, 2025). To contend with these perceived challenges, CCP leaders aim to bolster indigenous technological development, strengthen China’s military capabilities, align more closely with Russia and other U.S. adversaries, and strengthen China’s ties with countries in the Global South.

Given these sharply competitive dynamics and mutual threat perceptions, the United States and PRC are likely to face strong incentives for competition and perhaps even conflict in some areas. At the same time, however, their incentives might align for cooperation in other areas, and in others still there are likely to be mixed incentives for both competition and cooperation. The table summarizes the incentives for conflict, competition, and cooperation across each of the areas that RAND researchers Jim Mitre and Joel Predd have characterized as “AGI’s five hard national security problems” (Mitre and Predd, 2025).

Table 1. U.S.-PRC Incentives Across AGI’s Five Hard National Security Problems

In the remainder of this paper, we explore the mix of U.S. and PRC incentives for conflict, competition, and cooperation across each of these five areas.

Wonder Weapons
Imagine powerful AI that suddenly gives the United States or China a completely unexpected and potentially decisive weapon—for example, a missile defense system capable of negating an adversary’s nuclear deterrent, or a cyber capability to completely disrupt a country’s critical infrastructure. The emergence of such a “wonder weapon” could change the calculus of risk and deterrence for both nations. In an extreme case, a country possessing a wonder weapon might be tempted to use it for decisive strategic advantage; a nation without a wonder weapon might feel the necessity to strike first against an adversary nation about to achieve one.

The development of wonder weapons will likely accelerate as the United States and PRC engage in intensifying military competition, with both sides racing to achieve technological breakthroughs that could provide decisive advantage in future conflicts. Wonder weapons might also increase the risk of miscalculation and escalation because of opaque AI decisionmaking. Indeed, potential arms race dynamics may lead to strategic instability, making conflict between the two countries more likely. However, the potential disruption of deterrence from any wonder weapon, and the accelerated risks of conflict, may be a reason for cooperation, at least in terms of de-escalation and reassurance. Cooperation in the form of mutual agreements on restraint and transparency for advanced AI-enabled military technology will be extremely difficult, with both the United States and PRC racing to achieve potentially decisive advantage. Yet they could see some degree of mutual restraint as desirable. For example, it might be possible to build on the agreement that AI should not make decisions about the use of nuclear weapons, with the two countries considering how they might both benefit by reducing the risk that AI-enabled systems could cause accidental or inadvertent escalation.

Systemic Shifts in Power
Each side’s approach to the problem of systemic shifts in power will likely be driven by competition as their security, economic, and technological rivalry deepens. The U.S. military and the PLA will almost certainly compete to use AI in ways that could cause major shifts in aspects of military competition. Even if those uses fall short of being wonder weapons, shifts in military competition could still lead to intensifying competition, especially if one or both sides view the shifts as upending key aspects of the military balance between the two countries.

For example, the PRC could use advanced AI capabilities to enhance its external propaganda and political influence operations. The PLA is already showing interest in using advanced AI capabilities to bolster the effectiveness of its “cognitive domain warfare,” such as by creating and distributing increasingly realistic deepfakes and tailoring messaging to specific audiences (Beauchamp-Mustafaga, 2024). The PRC could use such AI-enabled capabilities as tools to manipulate international public opinion in its favor, erode public support for cooperation with the United States in key countries, or even to affect the outcome of elections in Taiwan or other democracies (Marcellino et al., 2023). Meanwhile, Beijing will likely fear that the United States or other countries could attempt to use their own AI-enabled capabilities in support of perceived efforts to undermine the control of the CCP domestically.

An additional point of competition is the potential for AI to confer major economic benefits and scientific breakthroughs, especially as security rivalry and economic and technological policies become more deeply intertwined. Generative AI agents have already begun to produce advances in fundamental computer science and, while estimates vary, are likely to bring about meaningful accelerations in economic growth (AlphaEvolve Team, 2025). If China or the United States were to more effectively diffuse AI in its economy and gain even a modest (e.g., 1 percent) improvement in gross domestic product, over time, this could be a decisive advantage for that nation.^[2] So while there may be potential below the state-to-state level for cooperation (e.g., at the level of business corporations) the national stakes for such an advantage are so high as to make competition likely.

There are, however, areas of potential cooperation. Scientific advances do not need to be solely competitive. For example, shared interests in medical and health advances could be the basis for cooperation between the states. Likewise, environmental science breakthroughs or advances in disaster response could have broad, global effects that warrant cooperation. To be sure, the increasingly sharp, competitive dynamics of the broader bilateral relationship have led to diminishing prospects for cooperation in these areas. Something would have to change, perhaps dramatically, for the two countries to pursue these areas of potential cooperation. Indeed, it might take something like an unprecedented global health crisis or natural disaster for the two countries to judge they have sufficient shared incentives to overcome barriers to cooperation.

Non-Experts Empowered to Develop Weapons of Mass Destruction
Advances in AI, particularly generative models, likely lower the barriers for non-experts to develop WMDs, including nuclear, chemical, and biological weapons. By lowering the technical knowledge barriers to entry, powerful AI might enable individuals or nonstate actors to develop WMD capabilities that were previously restricted to nation-states or highly specialized experts.

The proliferation of WMDs poses a shared threat to both the United States and the PRC, as neither nation wants to contend with the destabilizing risks of nonstate actors armed with such weapons. This mutual interest creates an opportunity for cooperation in such areas as monitoring AI applications, creating institutions for global intelligence-sharing about misuse of AI, establishing safeguards against misuse, and developing international norms to prevent the exploitation of AI for WMD development. Collaborative efforts could reduce global risks while fostering trust between the two nations.

At the same time, however, strategic distrust, competing strategic priorities, and reluctance to share sensitive intelligence capabilities could make cooperation more challenging. Moreover, this may be a space of competition, as well. China is currently positioning itself as the champion of open-source AI for the broader good of humanity, in contrast to proprietary AI development in the United States. So, while both might be open to cooperation over misuse of AI, soft power competition over “enabling misuse” through open-source AI might also emerge as a new area of friction.

Artificial Entities Developing Uncontrolled Agency
One of the most transformative aspects of current AI systems is their ability to exhibit agentic qualities: They can plan and execute tasks with a degree of autonomy, often without direct human supervision. For example, large language models can generate and execute code, and they can interact with external systems when connected to the internet or internet-connected hardware. These capabilities hold great promise for solving complex problems and driving innovation, but they also carry significant risks.

Even without achieving sentience or AGI, agentic AI systems could malfunction or act unpredictably, potentially causing widespread harm. For instance, imagine an AI-powered air traffic control system that, because of a design flaw or unforeseen circumstances, directs planes into collisions across the nation. Such scenarios highlight the critical need for robust safeguards and oversight mechanisms to ensure that these systems operate as intended. Both the United States and China will likely share incentives for cooperation when it comes to ensuring the control and safety of agentic AI systems.

Instability
When it comes to the last of the hard problems—instability—Washington and Beijing will face competitive dynamics and diverging incentives. To some degree, this may be an unavoidable outgrowth of great-power competition, but the situation could be exacerbated if each country views itself as pursuing stability and the other as taking provocative and potentially destabilizing steps that could upset a potentially fragile balance between them. For example, imagine a large U.S. investment in AI cybersecurity that Beijing, because of misperceived intentions, interprets as actually being about offensive strategic cyber capabilities and increasingly unstable dynamics between the two nations.

At the same, time, however, the United States and the PRC will likely also have incentives to engage in at least some level of cooperation. They may both wish to reduce the risk that their race to develop AGI will be so destabilizing that it leads to outcomes damaging to both countries’ interests. As RAND researchers Mitre and Predd (2025) note,

Whether AGI is ultimately realized or not, the pursuit of AGI could foster a period of instability, as nations and corporations race to achieve dominance in this transformative technology. This competition might lead to heightened tensions, reminiscent of the nuclear arms race, such that the quest for superiority risks precipitating, rather than deterring, conflict. In this precarious environment, nations’ perceptions of AGI’s feasibility and potential to confer a first-mover advantage could become as critical as the technology itself.

This raises the possibility that misperception, misinterpretation, or miscommunication could lead to outcomes that neither side desires. Despite intense security, economic, and scientific and technological rivalry, both countries have incentives to avoid inadvertent escalation or unintentional conflict that could be caused by pursuit of advanced AI capabilities.

Conclusion
For some national security problems related to the pursuit of AGI or advanced AI capabilities, the United States and the PRC will face strong incentives for competition, and possibly even for conflict. There is every reason to believe that these incentives for competition will deepen in the coming years, especially if tensions over Taiwan, maritime disputes in the South China Sea, or other security issues continue to increase. Yet, some of the other hard problems, such as the threat of non-experts empowered to develop WMDs and artificial entities with agency, appear to present a different set of incentives for Washington and Beijing as they navigate the path toward AGI. Moreover, in each of these areas, at least some degree of cooperation is likely to leave both parties better positioned to avoid undesirable outcomes—if they can manage to achieve it despite their intensifying competition and increasingly sharp differences. Additionally, the two sides would likely benefit from some discussions about topics related to the areas of intense competition, to reduce the risk that competition could spiral into conflict because of misperception, misunderstanding, or miscommunication.

The channels of communication required to cooperate where incentives align, and to reduce risks where they diverge, will likely take a long time to develop and institutionalize, and they could be difficult to maintain in the event of a further downturn in the U.S.-China relationship. Nonetheless, policymakers should focus not only on outcompeting the PRC where required, but also on creating and sustaining opportunities for cooperation where it is in the United States’ and the PRC’s interests. Even where the two countries’ interests in cooperation align, progress on official communications mechanisms is likely to be slow, at least absent a catalyzing event.

In the near term, the United States can start to build a foundation through a wide range of steps: Track 2 or Track 1.5 dialogues, convening expert working groups on terminology, and holding discussions on issues related to these problems as part of existing policy dialogues. To be sure, there are likely to be barriers to cooperation, including the need for military secrecy, the PRC’s long-standing reluctance to engage in strategic stability and arms control discussions with the United States, and enduring strategic mistrust between the United States and China. Moreover, the incentives for cooperation may not align until an event occurs. Yet there is an urgency now to build a foundation for future cooperation where possible (Predd, 2025), as well as for dialogue to reduce the risk that U.S.-China competition will spill over into an otherwise avoidable conflict that could be extraordinarily damaging for both countries.

Acknowledgments
We acknowledge Salman Ahmed, Jude Blanchette, Edmund J. Burke, and Gerard DiPippo for providing insightful reviews of earlier drafts, and Jim Mitre, Joel B. Predd, and Emma Borden for supporting the development and publication of this paper.

Notes

[1] We define AGI broadly here: AI systems with human or superhuman capabilities across a variety of tasks, or that can widely perform economically valuable work by substituting automation for labor, or that display emergent learning properties for new skills and tasks.

[2] The future is uncertain, and we can imagine multiple variations of economic advantage—for example, one nation being first to diffuse and adopt AI, with the second nation catching up. However, we consider the real possibility that AGI could provide enduring advantages that can be sustained.

References

·  AlphaEvolve Team, “AlphaEvolve: A Gemini-Powered Coding Agent for Designing Advanced Algorithms,” DeepMind, May 14, 2025. As of July 11, 2025:
https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/

·  Beauchamp-Mustafaga, Nathan, Exploring the Implications of Generative AI for Chinese Military Cyber-Enabled Influence Operations: Chinese Military Strategies, Capabilities, and Intent, RAND Corporation, CT-A3191-1, 2024. As of July 11, 2025:
https://www.rand.org/pubs/testimonies/CTA3191-1.html

·  Bradsher, Keith, “China’s Leader, with Rare Bluntness, Blames U.S. Containment for Troubles,” New York Times, March 7, 2023.

·  Colby, Elbridge A., The Strategy of Denial: American Defense in an Age of Great Power Conflict, Yale University Press, 2021.

·  Doshi, Rush, The Long Game: China’s Grand Strategy to Displace American Order, Oxford University Press, 2023.

·  Marcellino, William, Nathan Beauchamp-Mustafaga, Amanda Kerrigan, Lev Navarre Chao, and Jackson Smith, The Rise of Generative AI and the Coming Era of Social Media Manipulation 3.0: Next-Generation Chinese Astroturfing and Coping with Ubiquitous AI, RAND Corporation, PE-A26791-, 2023. As of July 11, 2025:
https://www.rand.org/pubs/perspectives/PEA2679-1.html

·  Mitre, Jim, and Joel B. Predd, Artificial General Intelligence’s Five Hard National Security Problems, RAND Corporation, PE-A3691-4, 2025. As of July 11, 2025:
https://www.rand.org/pubs/perspectives/PEA3691-4.html

·  Predd, Joel B., Riding the Coming Wave: The Urgent Need for Contingency Plans for AGI Futures, RAND Corporation, WR-A3691-1, 2025. As of July 11, 2025:
https://www.rand.org/pubs/working_papers/WRA3691-1.html

·  State Council Information Office of the People’s Republic of China, National Security in the New Era, white paper, May 12, 2025.

·  U.S. Embassy and Consulates in China, “Readout of President Joe Biden’s Meeting with President Xi Jinping of the People’s Republic of China,” November 17, 2024. As of July 11, 2025:
https://china.usembassy-china.org.cn/readout-of-president-joe-bidens-meeting-with-president-xi-jinping-of-the-peoples-republic-of-china-3/

Michael S. Chase is Senior Political Scientist at RAND. William Marcellino is Senior Behavioral and Social Scientist; Professor of Text Analytics, RAND School of Public Policy. This article is published courtesy of RAND. Funding: The Geopolitics of AGI Initiative is independently initiated and conducted within RAND’s Technology and Security Policy Center using income from operations and gifts from philanthropic supporters, or which have been recommended by these philanthropic supporters. A complete list of donors and funders is available at www.rand.org/TASP. RAND donors and grantors have no influence over research findings or recommendations.

Leave a comment