The Party's Gambit: How China's Worldview Shapes Its Pursuit of AI

by Shanshan Mei

On July 17, Xi Jinping gave the keynote at China’s World AI Conference in Shanghai, the first time a Chinese president had done so in person. He described AI as “a symphony of international cooperation,” announced a new China-based body to help shape global AI rules, promised AI training for developing countries, and stressed that the technology must remain safe, controllable, and under human oversight. Much of the speech presented China as open, responsible, and willing to cooperate internationally. At the same time, Xi made clear that Beijing intends not only to participate in global AI governance, but to shape it. The speech positioned China as seeking normative influence in AI while competing with the United States to define its future.

AI may be one of those rare technologies whose consequences are best measured at the scale of civilizations, not of markets or election cycles alone. Under those conditions, strategy will be shaped not only by capability, but also by worldview. For China, the key question is how the Chinese Communist Party (CCP), despite its rhetoric about a “shared future for mankind,” reconciles national interest with the broader geopolitical impact of AI. This essay argues that Beijing’s approach to advanced AI is shaped by two competing fears, the fear of disorder and the fear of falling behind, and by a related ambiguity in what the Party means by control itself. The CCP appears to see AI’s dangers and its strategic necessity at the same time, yet remains inclined to press ahead. Understanding that tension is essential to interpreting China’s behavior in the coming years.

This cuts against a comforting assumption in Western analysis. The Party’s desire for control is treated as overriding its drive for technological advance, and therefore as making Beijing more cautious or more open to limits. The Party’s own historical lessons point in the opposite direction. In its view, technological inferiority has itself produced instability, vulnerability, and strategic decline. For the CCP, control and technological advancement are not in tension. Technological advancement has become a condition of control.

The CCP’s approach to AI is shaped by two fears that point in opposite directions but arise from the same imperative: the survival of the Party-led order. One is the fear of disorder. The other is the fear of falling behind. Disorder threatens the regime from within; technological inferiority threatens it from without. Because both are tied to the same survival logic, neither can simply override the other. This tension helps explain why Beijing’s approach to advanced AI remains unsettled.

The Party’s fear of disorder has deep historical roots. Unlike a party that holds office for a term, the CCP presents itself as the custodian of a civilization that Xi Jinping calls the world’s “only great civilization to have continued unbroken and developed in state form to the present,” and it treats that continuity as both an accomplishment and a source of legitimacy. In this view, order is the normal condition of politics, disorder a deviation to be corrected, and the gravest danger a rupture deep enough to prevent recovery. The CCP inherited this outlook and recast it through a Leninist framework, linking the structure of a disciplined vanguard party to a civilizational mission. That instinct was reinforced by the Party’s own most traumatic experiences. The Anti-Rightist Campaign, the Great Leap Forward and the famine it produced, and the Cultural Revolution were not external shocks but internally generated crises, with immense human costs. Many of China’s current leaders, including Xi Jinping, came of age during those years, often in families that were themselves targets of persecution. The lesson the leadership drew was stark: uncontrolled upheaval, whatever its ideological banner, is among the gravest threats a regime can face.

The same history also shaped the Party’s view of science and technology and gave rise to a second fear. Mao-era campaigns repeatedly subordinated technical and scientific judgment to political loyalty, with disastrous results. The post-Mao leadership concluded that a state unable to manage the relationship between political authority and technical knowledge risks weakening itself from within. Deng Xiaoping made the reversal explicit at the 1978 National Science Conference, arguing that the persecution of intellectuals had served to “wreck the productive forces and our socialist construction” and insisting that scientific and technical workers were “part of the working class itself”—a deliberate effort to reintegrate technical knowledge into political authority. Deng later framed the issue in terms of national power, arguing in 1988 that without “the atomic bomb, the hydrogen bomb, and satellites,” China “could not be counted a major power of real influence.”

This helps explain why the modern CCP treats science and technology not as a neutral field of innovation but as a matter of sovereignty and political security. China’s national security doctrine makes the hierarchy explicit. The Overall National Security Outlook places political security—the security of the Party-led system itself—at the foundation of national security as a whole. That hierarchy was reaffirmed in the 2025 white paper on national security and elaborated by Minister of State Security Chen Yixin in a companion Qiushi essay on safeguarding modernization through “high-level security [高水平安全].” In this framework, falling behind technologically is not merely an economic liability. Read through the memory of the late Qing collapse before better-armed foreign powers, it becomes an existential vulnerability.

Taken together, these two fears define the Party’s basic approach to AI. If disorder is the central danger, then AI raises an obvious concern: a sufficiently capable system could escape reliable supervision or be used to undermine domestic control, including through cyberattacks on the surveillance and governance systems on which the Party depends. But if falling behind is the central danger, then AI appears as a strategic domain in which China cannot afford to lag, especially relative to the United States. In that case, the greater risk lies not in moving too quickly, but in moving too slowly. One fear points toward caution, control, and limits. The other points toward speed, scale, and competitive urgency.

Chinese policy rhetoric increasingly reflects both impulses. In June 2026, the State Council released a white paper, “More Just and Equitable Global Governance,” that treats AI as a priority governance domain and frames the deeper contest through rivalry, condemning “small yard, high fences,” “parallel systems,” and the overstretching of national security in language clearly aimed at U.S. export controls. On the safety side, it holds that AI governance must keep to a bottom line of safety and ethics, insists that weapons systems remain under human control, and warns against an AI arms race. The same duality ran through the briefing releasing it: Politburo member and Foreign Minister Wang Yi announced the launch of a World AI Cooperation Organization, while the National Development and Reform Commission’s deputy director Zhou Haibing pledged to manage risks and guard against AI safety harms.

The evidence so far suggests that when Chinese elites have to choose an emphasis, they tend to lean toward the second fear rather than the first. Opening the 2025 World AI Conference, Premier Li Qiang stressed openness and “intelligence equalization,” arguing that AI should be a global public good that benefits all rather than the exclusive advantage of a few, and he used the occasion to propose a world AI cooperation body. At the 2025 Paris AI summit, former Vice Foreign Minister Fu Ying advanced a similar logic, arguing in effect that catastrophic risk becomes more likely when states remain trapped in geopolitical rivalry and placing primary responsibility on Washington. The same pattern appears in semi-official analysis, which in China often tracks closely with the state’s broader line. A 2026 paper co-authored by a scholar at Tsinghua University's Center for International Security and Strategy, titled “The ‘Silicon Curtain’ Descends” (硅幕已至), argues that U.S. export controls have become instruments for preserving hegemony, pushing China toward a more self-reliant and “autonomous and controllable” path. As Matt Sheehan and Scott Singer have observed, Beijing increasingly speaks of AI safety and AI acceleration in the same register, without resolving the tension between them.

That unresolved tension matters because the two fears imply different policy responses. If the overriding danger is a system that no one can control, the rational response is to slow down and put effective controls and safety measures in place before capabilities advance further. If the overriding danger is losing ground to a rival, the rational response is to accelerate, on the assumption that the technology will develop regardless and is safer in one’s own hands than in someone else’s. The Party appears to hold both views at once. Neither its history nor its ideology settles which one should prevail. The answer instead depends on contingent judgments: how close Beijing believes dangerous capabilities to be, how much confidence it has in its ability to manage them, and whether it sees rival powers as racing ahead or exercising restraint. Those judgments can change. That is why China’s response to the most dangerous forms of AI remains genuinely open.

The tension described in the previous section reflects not only two competing fears, but also a deeper ambiguity in Beijing’s approach to AI: the meaning of control itself. The Party invokes control constantly, but not always in the same sense. At least three distinct ideas are at work: political control over society and institutions, strategic control over technological dependence and international rule-setting, and technical control over the behavior of advanced AI systems. These forms of control overlap, but they are not interchangeable. The Party’s confidence in one may obscure its limits in another.

For Beijing, AI is above all an instrument. It is expected to strengthen governance, improve economic performance and military might, and extend China’s influence abroad. In that sense, AI fits naturally within the Party’s governing logic. But that fit is also the source of the problem. The most advanced AI systems may not remain instruments in any ordinary sense.

The first kind is political control. The CCP’s governing reflex is to preserve authority over society, information, and institutions. AI is attractive to the CCP in this sense because it promises to make a large and complex society more legible, surface risks earlier, and strengthen anticipatory governance. Xi Jinping’s recurring warnings about “black swans” and “gray rhinos,” together with the injunction to remember danger in times of peace (居安思危), reflect a worldview in which risk accumulates quietly and stability depends on early intervention. The official record reflects this logic. At the October 2018 Politburo study session on AI, Xi folded the technology into “social governance,” called for guarding against its risks, and said it must remain “safe, reliable, and controllable.” By the April 2025 session, risk had moved much closer to the center. Xi described AI as a source of “unprecedented risks and challenges,” called for systems of technology monitoring, early warning, and emergency response, and linked controllability to self-reliance and strategic initiative. The vocabulary of control remained constant. What changed was the prominence of risk and the sharper competitive framing around it.

The second kind is strategic control. Here the concern is not domestic authority, but China’s position in the international system. The Party treats AI as a domain in which dependence on foreign suppliers, standards, or platforms would create strategic vulnerability. Calls for “autonomous and controllable” development, objections to “technology monopoly,” and opposition to U.S. export controls all reflect this concern. In this sense, control means reducing external dependence, preserving freedom of action, and shaping the rules under which others must operate. It is closely tied to technological leadership.

The third kind is technical control, and here the Party’s familiar vocabulary becomes less reliable. As the leadership often uses it, “controllable” is largely a political concept. It means that systems should remain subordinate to human authority, subject to oversight, and capable of being interrupted or redirected. This resembles the hierarchical control the Party seeks over media, markets, and bureaucracies. But the control required for highly capable AI systems is different. It cannot simply be asserted from above. It must be built into the system through design, evaluation, and testing, and it may fail precisely where political authority is strongest. In the AI-safety literature, the concern is not merely that a system might disobey instructions, but that a sufficiently capable system may not be reliably overseen at all. The problem is not just whether there is a lever, but whether the lever still works.

This distinction is not only theoretical. It is visible in Chinese discourse itself. The term kekong (可控), usually translated as “controllable,” is used both for political control and for the technical challenge of keeping AI systems bounded and corrigible. But the two meanings are beginning to diverge. Leadership rhetoric still points mainly to human authority, intervention, and override. China’s technical and standards communities, by contrast, increasingly use the same term to refer to alignment, safe termination, self-replication, and loss of control in the technical sense. The AI Safety Governance Framework 2.0, released in September 2025, made “preventing loss of control” an explicit objective and called for built-in safe termination mechanisms. TC260’s Ethics-Safety Guidelines for Artificial Intelligence Applications 1.0, issued the following May, warned against the risks posed by highly autonomous systems and by failures of human agency when systems act beyond what users anticipate and existing safeguards fall short. In other words, the gap between political and technical control is increasingly visible within China’s own system.

That gap cuts both ways. A regime that places unusual value on control may be more sensitive to catastrophic loss-of-control risks than outside observers assume. But a regime accustomed to imposing political control may also overestimate its ability to impose technical control. That is the central problem. Competence at controlling society is not the same as competence at controlling highly capable AI systems, and success in the first may breed false confidence in the second.

China’s history does not determine its approach to advanced AI, but it does shape the Party’s baseline assumptions. The CCP appears to view advanced AI through two fears that arise from the same priority of regime survival: the fear of disorder and the fear of falling behind. The first creates reasons to take loss-of-control risk seriously. The second creates reasons to continue moving quickly, particularly if Beijing believes the United States is doing the same. In the Party’s view, technological backwardness has itself been a source of political instability, national weakness, and strategic vulnerability. That is why control and technological advancement are not necessarily competing priorities in Beijing’s thinking. Technological primacy is increasingly understood as a condition of control. This does not mean that the Party will ignore loss-of-control risk. It means that such risk is unlikely, by itself, to outweigh the perceived dangers of falling behind.

Such an assessment makes competition over advanced AI harder to stabilize than it may appear. Beijing may speak the language of safety, controllability, and international cooperation, and it may in some cases take catastrophic risk seriously. But its underlying incentives are unlikely to favor meaningful restraint if restraint is seen as weakening China’s strategic position. The Party’s experience in political and strategic control may also reinforce confidence that advanced AI can be managed through existing institutions, even where technical control remains uncertain. The result is not a China that is simply reckless or simply cautious. It is a China for which AI leadership is increasingly tied to regime security and national power. That linkage should temper any expectation that the CCP’s preference for control will, by itself, make Beijing a reliable partner in reducing frontier AI risk.