ChatGPT as a security threat: US–China security dilemma in the generative AI race

Abstract

The rise of AI has introduced new complexities into the traditional US–China security dilemma, with both sides viewing each other’s technological advances as potential threats. This article examines how ChatGPT has further complicated these dynamics. Its success triggered not only Chinese anxiety over losing ground in technological and strategic competition but also concerns about threats to ideological and regime security. In response, China implemented a series of measures aimed at regulating generative AI to preserve domestic stability while simultaneously advancing its own technological capabilities. These reactions, coupled with the technical success of China’s DeepSeek, reshaped the competitive landscape – this time triggering anxiety and strategic insecurity in the United States and propelling the US–China generative AI race into a new, more intense phase. This article focuses on China’s initial security concerns about ChatGPT, shedding light on the origins of the AI race and how AI has become deeply entangled in broader security dynamics.

Keywords

AI AI race ChatGPT DeepSeek generative AI security competition security dilemma strategic competition US-China relations

In the context of the intensified US-China ideological battle, the United States (US), leveraging its technological hegemony, has enabled certain generative AI systems to use ‘algorithmic black boxes’ to deliver anti-China content, thus shaping international audiences’ perceptions of China in a negative light. Generative AI creates content that can subtly infiltrate China’s ideological security, allowing the U.S. and the West to construct a multi-dimensional, comprehensive, and multi-perspective battlefield of ideological influence and attack against China. This strategy seeks to erode, dilute, and undermine the Chinese Communist Party’s (CCP) and the Chinese government’s control and influence in both domestic and international ideological domains, accelerating the process of ‘peaceful evolution’ within China.

- Excerpt from a Chinese article published by National Security Studies (a Chinese academic journal) (Xie & Cao, 2024)

Introduction

The rapid development of AI as a dual-use technology has introduced new complexities to US-China strategic competition by blurring the line between civilian and military domains and accelerating the pace of competition. In a traditional security dilemma, where Washington and Beijing remain uncertain about each other’s strategic intentions, AI investments and advancements – regardless of whether they serve defensive, offensive, or civilian purposes – are frequently perceived as potential threats. This mutual suspicion fuels insecurity and reinforces the competitive dynamics between the two powers. This article argues that the logic of the security dilemma applies to the case of ChatGPT. It examines how these dynamics unfold in the context of generative AI, focusing on ChatGPT’s catalytic role in accelerating – if not initiating – the generative AI race between the US and China.

The success of ChatGPT sparked worldwide excitement and surprise in 2023. Its exceptional ability to generate textual content and ideas has made it the fastest-growing consumer software application in history. Despite its global popularity, ChatGPT has faced considerable resistance – countries ranging from Russia, North Korea, and Cuba to Italy have banned access for various reasons. In China, ChatGPT has triggered a series of policy responses – including access restrictions and security measures – shaped by the broader context of US–China strategic competition. These responses have been driven not only by significant strategic anxiety but also by a deepening sense of insecurity. Many in China view generative AI as both a technological challenge and a national security imperative. This has prompted urgent efforts not only to close the innovation gap with the US, but also to safeguard its own security.

More specifically, ChatGPT is seen as posing not only universal social and legal challenges but also unique ideological and regime security threats to China. As a powerful content generator rooted in Western liberal values, ChatGPT is perceived as having the potential to produce information unfavourable or biased towards China. As the opening quote suggests, this, in turn, could bolster rivals’ ability to manipulate both international and domestic public opinion to China’s disadvantage, threatening its ideological security and fostering a hostile external environment. As such, China swiftly introduced regulations, becoming the first country in the world to establish a regulatory framework for generative AI.

Interestingly, these swift regulatory efforts by China have, in turn, sparked concerns in the US about China gaining a first-mover advantage in shaping global AI rules and norms, as well as posing a potential challenge to American leadership in AI development. The later technical breakthrough by China with DeepSeek further shocked the world, especially many in the US. As President Trump remarked, ‘The release of DeepSeek, an AI from a Chinese company, should be a wake-up call for our industries – we need to be laser-focused on competing to win’ (SkyNews, 2025). DeepSeek raised concerns in the US about China’s potential to influence American public opinion, elections, and democratic institutions more broadly. This led to a mirrored sense of insecurity regarding ideological and national security. As a result, it further accelerated the generative AI race between China and the US and intensified the dynamics of the AI security dilemma.

This article traces the origins of the US–China generative AI security dilemma by examining China’s initial perceptions of ChatGPT and the underlying logic that shaped its immediate policy responses. These early responses mark the beginning of a reactive chain in the generative AI race between the two powers, shedding light on how a downward spiral characteristic of a security dilemma can emerge – and how AI acceleration unfolds from the outset.

The remainder of the article is organised into several sections. The first reviews the literature on the US–China security dilemma and explores the new dynamics introduced by AI. The second outlines the study’s methodology and data sources. The third analyses how ChatGPT is perceived in China as a security threat – particularly in relation to ideological and regime security, as well as a source of asymmetric technological dependence – and examines the policy responses that have emerged from these perceptions. The final section concludes by summarising the key findings and offering policy recommendations aimed at easing security tensions.

AI and US–China security dilemma

Nowadays, the US and China are currently engaged in an intensifying strategic competition for global influence, technological leadership, economic power, and normative dominance in international politics. AI has emerged as a critical arena in this rivalry. As a transformative dual-use technology, AI introduces new variables and complexities into the strategic calculations of states. The speed, scale, and nature of AI capabilities have amplified uncertainties in threat perception and strategic decision-making. As a result, AI not only accelerates broader geopolitical competition but also intensifies security rivalry, as each side increasingly perceives the other’s AI advancements as a potential threat. This dynamic reflects the logic of the security dilemma – one of the most enduring theories in international relations – first articulated by Butterfield, Herz, and Jervis (Butterfield, 1951; Herz, 1951; Jervis, 1976, 1978).

Despite being used in contradictory ways in the international relations literature (Booth and Wheeler, 2008; Tang, 2009), the traditional security dilemma is often considered to arise within an anarchic international system, where states are inherently uncertain regarding the intentions of their counterparts, especially rival ones (Copeland, 2000; Mearsheimer, 2014a; Waltz, 1979). As such, actions taken by one state to enhance its own security are perceived as threats by others, particularly rival states. This dynamic has led to a cycle of mutual distrust, arms races, and potentially direct military conflict (Booth and Wheeler, 2008; Booth et al., 2011; Glaser, 1997; Jervis, 1978; Tang, 2009).

Contemporary US–China relations are one of the most widely discussed case studies of this security dilemma (Breuer and Johnston, 2019; Christensen, 1999; Kertzer et al., 2024; Steinberg and O’Hanlon, 2014). As growing geopolitical tension fuels mistrust and zero-sum interest narratives in Beijing and Washington, the efforts of either party, such as enhancing military capabilities in East Asia even with defensive aims, are perceived by the other party as aggression and thus inadvertently cultivate a sense of insecurity, prompting reciprocal security measures. These dynamics fuel a downward spiral of mounting tensions and hostility between China and the US.

Some scholars argue that China’s behaviour is driven by intentional revisionism, strategic ambition, and dissatisfaction with the liberal international order, rendering US–China conflict more inevitable and structural (Beckley and Brands, 2022; Friedberg, 2012; Mearsheimer, 2014a, 2014b). From this perspective, deterrence and containment – not reassurance or trust-building – are seen as the appropriate responses, and the security dilemma is dismissed as an inadequate explanatory framework. However, this ‘China-is-revisionist’ thesis often risks oversimplifying Chinese intentions, overlooking mutual threat perceptions, and neglecting the structural dynamics of misperception and escalation (Christensen, 2006; Johnston, 2019). Even rational, status quo-oriented states can descend into rivalry when uncertainty and mistrust dominate their strategic interactions – a dynamic that lies at the core of the security dilemma.

Needless to say, this US–China security dilemma is deeply concerning, as, in the worst-case scenario, a downward spiral in bilateral relations could lead to direct security conflict between the two largest world economies and nuclear states, bringing disastrous consequences for global security and prosperity. How to mitigate this US–China security dilemma and maintain peaceful bilateral relations has become one of, if not the most, important foreign policy questions of the 21st century, drawing considerable attention in the academic literature (Allison, 2015; Christensen, 2006; Mearsheimer, 2014a; Steinberg and O’Hanlon, 2014).

Nowadays, the rapid rise of AI has made it a central theme in US–China strategic competition. While a growing body of scholarship examines the politics of AI in the Chinese context (Cheng and Zeng, 2023; Ding, 2018, 2019; Qiao-Franco and Bode, 2023; Sheehan, 2018; Zeng, 2020, 2021a, 2021b, 2022, 2025; Zhang, 2025), the fast-paced development of AI and its increasing strategic significance mean that its specific interaction with US–China security dynamics remains underexplored.

Johnson’s work represents one of the early pioneering studies examining how AI contributes to uncertainties and insecurities within US–China security competition (Johnson, 2023). His research highlights how both the structural and non-structural characteristics of AI technologies intensify the security dilemma between the two powers. This article further applies the security dilemma framework to examine the dynamics of the US–China AI race, with a particular focus on the origins of the generative AI race. It analyses China’s security concerns about ChatGPT and the subsequent chain of reactive measures that intensified strategic competition.

It is important to reconsider the traditional security dilemma by incorporating the added complexity and uncertainties introduced by AI. As one of the most powerful emerging technologies, AI has considerable potential, sparking ongoing debates. However, these debates occur in a context where the future development of AI and its potential impact remain highly speculative at the time of writing. Compared to other transformative technologies, AI stands out due to its unique ability for self-amplification and uncontrollable diffusion (Suleyman and Bhaskar, 2023). This dynamic nature grants it unpredictable trajectories and potentially limitless capabilities. The inherent uncertainty surrounding AI’s true potential leaves ample room for misinterpretation, complicating security dilemma dynamics in international relations. In addition, AI’s actual capabilities in areas such as military domains (e.g. nuclear weapon systems) are often highly classified and may not be available to shape the perceptions currently influencing actual policies. In these regards, assessments of AI development are often based on uncertainties and unknowns, complicating the security dilemma.

Moreover, like space, cyber, and nuclear technologies, AI’s dual-use characteristics have created additional uncertainties in the security dilemma. The traditional security dilemma often categorises states’ security measures into a binary distinction: defensive versus offensive. This binary distinction, however, does not always adequately capture the dynamics of dual-use technologies like AI, as it is not explicitly designed for traditional military purposes – whether defence or offence – even though its advancements in civilian sectors can quickly translate into military capabilities. For example, AlphaGo was not designed for military purposes. However, its success has drawn worldwide attention to its military potential and led to considerable strategic anxieties within the Chinese military community (Hu, 2016; Tao et al., 2016; Zeng, 2025).

ChatGPT is no exception to this military utility perspective. Indeed, while OpenAI initially banned the use of ChatGPT for military purposes – a position that slightly shifted at a later stage – the rise of generative AI has inevitably drawn worldwide attention to its implications for military use from the very beginning. The Chinese military community has been closely observing the impact of ChatGPT on elevating American military capabilities. One headline from a Chinese newspaper, for example, cautions about the growing collaboration between the US military and its tech giants, warning against the potential militarisation of generative AI. The article highlights the role of large language models in not only collecting and analysing vast amounts of military intelligence but also in assisting decision-making (GT, 2024).

There is also a higher risk of fake information produced by generative AI, which could mislead military command and control systems (GT, 2024). Some recognise ChatGPT’s potential to influence human cognition, noting how the rise of ‘cognitive weaponization’ has enabled ‘AI-driven cognitive warfare’, posing a threat to China’s national security (Zeng and Mao, 2024). In this regard, the dual-use nature of AI has introduced additional complexities and uncertainties in estimating adversaries’ intentions, motives, and capacities.

These key technical characteristics of AI have further complicated the US–China security dilemma and heightened mutual concerns. According to Robert Jervis’ classical analysis, offence-defence balance and offence-defence distinguishability are two key variables that regulate the intensity of the security dilemma (Jervis, 1978). While its precise impact remains uncertain and it can serve both offensive and defensive purposes (Garfinkel and Dafoe, 2019), generative AI is widely regarded as shifting the offence–defence balance in favour of the offence (Bonfanti, 2022). Its speed, scalability, and low deployment cost enable actors to conduct information manipulation, cyber intrusions, and influence operations more efficiently than traditional tools, thereby increasing the perceived advantages of pre-emption (Goldstein et al., 2023; Marcellino et al., 2023).

As for offence-defence distinguishability, generative AI presents significant challenges in the cyber domain. As mentioned, its dual-use nature makes it difficult to distinguish between offensive and defensive or civilian applications. The same language model can be used for benign purposes such as translation, education, and healthcare, as well as for malign activities like deepfake propaganda and automated cyberattacks – helping to explain the aforementioned Chinese assessments of ChatGPT’s potential military utility. This opacity of intent exacerbates uncertainty and encourages worst-case assumptions, thereby intensifying the security dilemma.

This combination reflects the most dangerous scenario in Jervis’ framework – a world where offence has the advantage and military postures are indistinguishable. Generative AI, therefore, heightens insecurity by simultaneously increasing the incentives for preemption and reducing the ability to signal peaceful intent, thus deepening the intensity of the security dilemma.

Building on Jervis and others, Tang Shiping’s influential work introduces additional material factors that regulate the intensity of the security dilemma, including asymmetric power, external actors, and ethnic group dynamics (Tang, 2009). He also suggests to revisit psychological factors, such as fear, ‘general aversion to loss’, and domestic politics, which can shape how states interpret and respond to potential threats (Tang, 2009). Fear and asymmetric power are particularly relevant to the generative AI security dilemma examined in this study.

As this article will discuss, there is growing fear in China that ChatGPT and other US-developed generative AI tools may serve as unfiltered channels for Western information and liberal values, potentially undermining the social and political order, and threatening regime security in China. Conversely, despite contrasting political systems, a similar fear is mirrored in the US, where many worry that Chinese-developed generative AI could be used to influence public opinion, interfere in democratic processes, and exploit societal divisions – thereby weakening American democracy and advancing authoritarian values that challenge the liberal international order more broadly.

These Chinese and American fears are also closely tied to asymmetric power dynamics, where one side’s material or technological superiority can significantly intensify the security dilemma. The weaker party often assumes the stronger will eventually use its advantage coercively, while the stronger power may fear that any relative decline will embolden the weaker actor to challenge the status quo. In the case of generative AI, as will be discussed, many in China view American dominance – exemplified by the success of ChatGPT – as reinforcing China’s reliance on a US-developed AI ecosystem, thereby exposing its growing strategic vulnerability. This asymmetry also reinforces the aforementioned fears regarding American discourse power and control over information flows – factors considered essential to China’s regime security. ChatGPT is perceived as accelerating this imbalance by shaping global narratives and enabling the spread of Western values that may undermine China’s ideological control.

On the US side, there are concerns that China’s ambitious generative AI initiatives could narrow – if not surpass – the current innovation gap, thereby challenging American AI superiority and shifting the balance of technological power between the two countries. In other words, generative AI is perceived as a powerful force capable of altering asymmetric power dynamics, with both sides fearing that the other’s technological edge could translate into geopolitical and ideological advantage.

Needless to say, these security dilemma dynamics and the generative AI race triggered by ChatGPT have emerged within the context of already heightened geopolitical tensions between the US and China, where AI has become a central theme in their broader strategic competition. In this environment, each side increasingly perceives the other’s AI progress as a potential threat to its own national security. For example, China’s AI development is perceived by many in the US as serving harmful applications or hegemonic purposes (NSCAI, 2021). From an AI-enabled surveillance state repressing civil rights to AI-enabled technologies manipulating democratic elections (NSCAI, 2021), these narratives – rightly or wrongly – perceive China’s AI advancements as a direct threat to the US, thus creating a sense of insecurity.

Similarly, the Chinese government has been framing AI as a matter of national security in its official discourse (Zeng, 2021b). From American AI plans to its major technological breakthroughs, China has been closely observing these developments and reflecting on its own progress (Zeng, 2025). China’s fear about ChatGPT must be understood within the broader context of intensifying US–China geopolitical tension.

Method and data

To examine how ChatGPT is framed as a security issue in China, this study draws on open-source Chinese materials – including policy documents, media reports, and academic publications. Content analysis is employed to assess relevant sources, with a particular focus on the most cited and most downloaded articles in the China National Knowledge Infrastructure (CNKI) database. These Chinese-language materials provide direct insight into how officials, scholars, and media outlets conceptualise generative AI. While internal deliberations behind key policy decisions remain opaque and some media content serves a propagandistic function, such sources nonetheless offer valuable indications of prevailing narratives and strategic thinking.

To mitigate the risk of misinterpretation that comes with relying solely on open-source materials, the analysis is supplemented by fieldwork involving multiple trips to China between 2024 and 2025. During this period, the author conducted interviews and informal exchanges with dozens of Chinese scholars from diverse backgrounds in politics and technology policy, as well as with institutions such as the Institute for AI International Governance (I-AIIG). Some of these conversations took place during conferences, including the 2025 World Artificial Intelligence Conference (WAIC) and other AI-related dialogues. In addition, the author presented preliminary findings at various conferences and guest talks, including several held in China and practitioner-oriented forums.

Fear, ideological, and regime insecurity

Like many emerging technologies, AI has the potential to disrupt social and political order. From a regime perspective, it is not surprising that ChatGPT is viewed as a potential threat to political stability, prompting the need for mitigation strategies to safeguard regime security. As a powerful content and idea generator, ChatGPT has sparked a sense of ideological insecurity in China. Its ability to influence the flow of information is perceived as a threat by many authoritarian regimes, such as North Korea, Iran, Russia, Cuba, and Syria, all of which have banned access to ChatGPT.

To put this perception of threat and insecurity into perspective, the rise of AI has also raised ideological concerns in Western liberal democracies. China and Russia are often cited as state actors capable of using AI to manipulate Western domestic elections and achieve their geopolitical goals. Through deep fakes, fake news, and biased algorithms, American intelligence agencies argue that AI has expanded authoritarian countries’ toolkit to undermine liberal democracies and their core values (De Luce and Grumbach, 2024). Similar concerns about foreign interference are shared by many authoritarian regimes. In this regard, the anxieties surrounding generative AI are not unique to any particular regime type, but arise from concerns over the empowerment of rival states – making the security dilemma particularly relevant in this context.

In the Chinese context, the most direct concern is the potential impact on America’s ability to intervene in Chinese domestic politics. Some scholars, for example, worry about how Chinese public opinion could be manipulated by generative AI through mechanisms such as fake news, polarisation, and the creation of information cocoons (Chaoyang, 2023; Huang and Yao, 2024; Ji and Yan, 2023; Su and Wang, 2023; Wang, 2024a, 2024b; Xie and Cao, 2024). Soon after ChatGPT was released, some Chinese netizens were investigated for using it to produce fake news (Zhuang, 2023). This has fueled concerns that Western countries could leverage generative AI to influence China’s public opinion (Chaoyang, 2023; Pu and Xiang, 2023). To some, similar to social media platforms like Twitter (now X), generative AI further enables the US to intervene in the public opinion of other countries or even subvert their regimes. It diversifies the American toolkit for conducting what some view as ‘ideological infiltration’ (Huang and Yao, 2023). As such, there is growing alarm over the increasing capacity of the US and the West to undermine or even overthrow regimes like China’s, thanks to the capabilities of generative AI (Wang, 2024b; Xie and Cao, 2024).

To many, it is not just Chinese public opinion at stake. Generative AI is seen as further empowering Western states’ ability to influence global attitudes, potentially creating an unfavourable external environment for China (Chaoyang, 2023; Ji and Yan, 2023). Some argue that Western countries, by leveraging their technological advantage in AI, can ‘strengthen their control over international public opinion’ and provide ‘more conducive ways to manipulate public opinion based on their value orientation’ (Huang and Yao, 2023).

For example, according to China Daily, ChatGPT’s responses on the topic of Xinjiang are consistently aligned with the American government’s political narrative that frames the situation as a ‘genocide’ in Xinjiang (China, 2023). This raises concerns that the US government can use generative AI to disseminate false information and manipulate global public opinion for its own geopolitical advantage (China, 2023). In this sense, generative AI is viewed as a powerful tool that significantly amplifies the influence of Western countries, particularly the US, and could have a ‘huge subversive impact on world politics’ (Gao, 2023).

While similar concerns have been raised regarding the rise of other information and communication technologies, such as social media, many argue that ChatGPT presents a unique challenge due to its ability to self-generate information that can influence public opinion. It is often pointed out that the corpus on which ChatGPT is trained is primarily based on Western values and thinking (Pu and Xiang, 2023; Shang, 2023; Song, 2025; Xie and Cao, 2024), with relatively limited inclusion of Chinese materials (Pu and Xiang, 2023; Xie and Cao, 2024). As a result, the content it generates tends to ‘cater to Western positions and preferences’ (Huang and Yao, 2023; Shang, 2023), reflecting a strong ideological bias (China, 2023; Xie and Cao, 2024).

Some critics argue that this bias is not incidental but rather an inevitable consequence of both the design of ChatGPT and the data it is trained on, thus reinforcing Western perspectives and potentially undermining alternative viewpoints (Shang, 2023; Xie and Cao, 2024). In this context, AI is increasingly seen as a tool of ‘ideological colonization’ and ‘cultural hegemony’—a new ‘ideological machine’ and ‘imperialist instrument’ that transforms virtual space into a key battleground for ideological competition among major powers (Wang, 2024b).

In the age of generative AI, differentiating accurate information is a common challenge for all societies. However, it is argued that generative AI presents unique challenges to China, particularly because its government prioritises ideological control and propaganda. In this context, ChatGPT is anticipated to complicate China’s existing efforts at maintaining ideological control (Shang, 2023; Wang, 2024b; Xie and Cao, 2024). One key issue cited is that the government’s resources for controlling information are limited in comparison with the vast and limitless amount of content generated by AI. The sheer volume and highly anthropomorphic nature of this content place significant pressure on the government’s ability to monitor, censor, and screen information effectively. As generative AI continues to evolve and proliferate, the government may find it increasingly difficult to maintain the same level of control over the information environment, potentially undermining its authority and ideological narrative (Xie and Cao, 2024).

Given ChatGPT’s potential to influence public opinion, some Chinese scholars foresee it as a threat that could ‘shake the ideological security’ of the country, posing new national security risks (Ji and Yan, 2023; Shang, 2023). Consequently, there have been calls for immediate action to safeguard China’s ideological security. These calls emphasise the need to ‘strengthen its mainstream ideology and its promotion mechanisms’ while also advocating for the construction of a regulatory framework specifically designed for generative AI (Wang, 2024b; Xie and Cao, 2024). Such measures are seen as necessary to prevent the erosion of ideological control and ensure that AI technologies align with the government’s values and priorities. This reflects the broader concern that generative AI, by facilitating the free flow of information, could challenge the centralised authority that underpins China’s governance.

While not traditionally classified as an ‘act of war’, the manipulation of public opinion by a foreign actor has increasingly been seen as a significant component of modern conflict. Considering the success of ChatGPT, some have raised concerns about the rise of AI-driven cognitive warfare, which relies on large language models (Chaoyang, 2023; Ji and Yan, 2023; Zeng and Mao, 2024). ChatGPT’s pro-American narratives are argued to amplify this form of cognitive warfare by shaping audience perceptions and behaviours through targeted information manipulation (Zeng and Mao, 2024).

It is argued that generative AI, by exploiting human cognitive vulnerabilities and biases, can induce specific patterns of thinking, leading to cognitive manipulation. This ‘cognitive weaponization’ is seen as a significant threat to national security (Zeng and Mao, 2024). In this context, generative AI has the potential to redefine traditional understandings of warfare and security conflicts, moving beyond conventional physical battles to encompass ideological and informational domains. The implications of this shift are profound, as it introduces new dimensions of conflict, where information control and public opinion manipulation become central to the strategies of both state and non-state actors.

Reactive chain: Competing to govern

To address the ideological and political risks posed by ChatGPT and other generative AI technologies, the Chinese government introduced interim Generative AI Measures, aiming to ‘promote Generative AI development’ while also ‘protecting national security’ (CAC, 2023). The law is based on the principle of ‘balancing development with security, promoting innovation while governing according to the law’. This legislation marks the world’s first comprehensive framework to regulate generative AI. According to the measures, generative AI needs to

adhere to the core socialist values and refrain from generating content prohibited by laws and administrative regulations, such as inciting subversion of state power, overthrowing the socialist system, endangering national security and interests, damaging the national image, inciting secession, undermining national unity and social stability, promoting terrorism, extremism, ethnic hatred, ethnic discrimination, violence, pornography, and false or harmful information. (CAC, 2023)

These provisions reflect China’s underlying concerns about the potential disruption generative AI may pose to its social and political order, as well as the rationale behind its policy responses. Although China’s concerns reflect widely shared global apprehensions about the risks of AI, its regulatory actions are not perceived as neutral. They are often interpreted through a geopolitical lens – particularly in the US – given the broader context of US–China security competition.

Some observers worry that China’s early regulatory efforts could grant it a first-mover advantage in shaping the global rules and norms governing AI, exacerbating anxieties about the perceived lack of equivalent action on the US side. For example, it is argued that the US ‘cannot afford to sit on the sidelines while China and Europe decide these fundamental issues for the world’ (Bradford, 2023). Senate leader Schumer has described China’s generative AI regulations as ‘a wake-up call to the nation’, emphasising that ‘it is imperative for the U.S. to lead and shape the rules governing such a transformative technology and not permit China to lead on innovation or write the rules of the road’ (Shepardson, 2023).

In other words, many in the US view China’s regulatory actions as part of a broader strategic effort to shape the normative and technical standards of global AI governance – posing a potential threat to the US. AI supremacy and national security. These concerns have, in turn, contributed to the development of emerging US regulatory efforts on generative AI. In this sense, ChatGPT has intensified US–China AI dynamics, swiftly transforming their strategic rivalry into a race to shape the development, deployment, and regulation of generative AI.

Strategic vulnerability: Technological asymmetry and dependence

In addition to ideological concerns, ChatGPT is widely perceived as reinforcing the technological asymmetry between the US and China, heightening concerns in China about growing technological dependence on US-based AI infrastructure and models. This perception has triggered anxiety among Chinese policymakers and scholars, who worry that such dependence could compromise information security and digital sovereignty.¹

It has been argued that generative AI could significantly enhance the capabilities of rival states – particularly the US – in intelligence gathering, narrative shaping, and strategic influence operations targeting China. These concerns contribute to a broader sense of strategic vulnerability and underscore the framing of generative AI as a security issue within China’s national security discourse. The argument is that the US could leverage AI technologies, like ChatGPT, to gather critical intelligence on other countries, including sensitive military, economic, and social data (Huang and Yao, 2023; Liu, 2023a). This would enhance its ability to monitor the moves of other nations, thereby eroding their information security and digital sovereignty.

Without effective countermeasures, it is argued that technologically weaker countries may become disproportionately transparent to technologically stronger ones, giving the latter strategic advantages. This technological asymmetry may engender an imbalance in global power structures, wherein technologically advanced nations leverage their superiority to conduct intelligence operations with relative impunity, perpetuating systemic vulnerabilities that leave less developed states incapable of safeguarding critical data assets and maintaining true ‘data security’ (Huang and Yao, 2023).

The widening technological gap is argued to increase the dependence of technologically weaker countries on technologically stronger ones (Liu, 2023a). In this dynamic, the latter can exploit their technological dominance, potentially forming a ‘monopoly and hegemony’ in critical technological sectors. This situation gives rise to a form of ‘new hegemony’ (Liu, 2023b), where the leading technological powers exert disproportionate influence over global affairs.

Some argue that the rise of generative AI will further intensify shifts in global power dynamics by exacerbating the imbalance of power between countries (Huang and Yao, 2023). As AI technologies like ChatGPT strengthen the positions of technologically advanced states, weaker nations may become more vulnerable to manipulation and control. This growing divide could undermine the sovereignty of less advanced nations, leaving them increasingly reliant on the technological hegemony of the more powerful states.

In response to this threat, many advocate for China’s technology independence and self-reliance (Pu and Xiang, 2023; Zhu and Yang, 2023). This call for self-sufficiency is framed as essential to countering the risks associated with the global dominance of AI technologies. By developing its own technological capabilities, China aims to reduce its dependency on foreign AI models and safeguard its national security, economic interests, and digital sovereignty.

This call for technology independence and self-reliance sits within the broader context of US–China technological competition, where China views mastering key technologies as essential to national security. The reliance on foreign technology is increasingly seen as a significant security risk (Webster et al., 2017). From the Trump administration’s decision to cut Chinese companies like Huawei and ZTE off from global semiconductor suppliers, to the Biden administration’s restrictions on China’s access to AI chips, American sanctions have amplified China’s resolve to achieve technological autonomy.

As Xi Jinping has articulated, ‘accelerating the development of a new generation of AI is an important strategic handhold for China to gain the initiative in global science and technology competition . . . We need to ensure that the core AI technologies are firmly in our own hands’ (China, 2019). This vision emphasises the necessity of achieving self-sufficiency in critical technologies, particularly AI, to safeguard China’s future technological and strategic interests.

In this context, the push to develop a domestic version of ChatGPT reflects China’s growing determination to reduce its dependence on foreign technologies, especially in key areas such as AI. As China seeks to protect its national security from potential threats posed by foreign technological influence, the desire for self-reliance also signals a broader trend of de-coupling the American and Chinese technology ecosystems. This de-coupling, spurred by both technological rivalry and geopolitical tensions, may redefine the global technology landscape and increase the divide between the two countries’ technological spheres of influence.

Reactive chain: Technological advancement and the pursuit of self-reliance

The success of ChatGPT highlights the rapid advancements of American AI innovation, casting China’s AI developments into the shadow. As mentioned, it has heightened China’s concerns about its technological competitiveness, particularly regarding the widening gap between China and the US. China was surprised by the potential of generative AI and began actively exploring how to develop its own generative AI industry. This led to a surge of enthusiasm among Chinese tech giants and start-ups, who quickly redirected resources to create ChatGPT-like language models, aiming to keep pace with American advancements and capture the Chinese market from which ChatGPT is banned.

In this context, when balancing the regulatory needs, the Chinese government was cautious about tightening control too much, as doing so could limit China’s AI innovation potential in the long run and grant the US a strategic advantage in the US–China AI race. These considerations of balancing short-term regulatory needs with long-term technological competitiveness framed China’s immediate response. This included introducing the aforementioned world first generative AI regulation aimed at ‘balancing security and development’ (CAC, 2023), encouraging domestic generative AI innovation, and accelerating the end of its crackdown on China’s tech giants.

Not surprisingly, in the context of intensified US–China strategic competition, China’s responses to ChatGPT quickly prompted reactions in the US. The breakthrough of China’s DeepSeek in particular signalled the country’s growing innovation capacity and technological competitiveness, effectively launching a full-scale generative AI race in innovation. This development fueled strategic anxiety and a sense of insecurity in the US. It also marked a turning point in the generative AI race, effectively shifting roles: the US found itself in a position of reflection, grappling with the implications of China’s rapid advancements and reassessing its own strategic posture in response.

DeepSeek’s emergence not only underscored the growing technological rivalry between China and the US but also heightened American fears over data privacy, potential foreign surveillance, and the stability of the American tech leadership in an increasingly competitive global landscape. These security concerns have led to a series of ongoing measures in the US aimed at safeguarding national security and economic interests – measures that reflect many of China’s initial anxieties and responses – thereby accelerating the US–China race in generative AI.

Concluding remarks

The rise of generative AI has introduced new dynamics to the US-China security competition. The emergence of US-developed ChatGPT triggered a wave of national reflection in China marked by fear over security and asymmetric technological power relations with the US. China’s strategic anxiety and security concerns have prompted a series of immediate responses aimed at regulating generative AI and bolstering its domestic AI industry. These responses, in turn, have sparked anxiety and insecurity in the US, not only regarding China’s ambitious plans in generative AI but also its regulatory efforts. In this context, ChatGPT has not only initiated the generative AI race but also contributed to a downward spiral of security competition and heightened geopolitical tensions.

Linking to the special issue’s three interconnected sections— ‘why’, ‘how’, and ‘so what’ regarding US–China competition – this article analyses how competition emerges in new domains, such as generative AI, accelerates into a race, and sets the stage for a downward spiral in the security dilemma. It also examines the Chinese context to understand the rationale behind China’s initial reactions to ChatGPT, shedding light on why this generative AI race began. Aligning with this special issue’s objective of bridging scholarly research and policy development, this analysis offers a modest policy-relevant insight that may help ease security tensions associated with the generative AI race, while acknowledging that the underlying dynamics of the security dilemma are unlikely to be fully resolved.

The potential, complexity, and uncertainties introduced by AI pose a serious risk of escalating tensions between the US and China. In the case of generative AI, it possesses the potential to amplify cognitive or psychological warfare.² While manipulation of public opinion by a foreign actor is often considered a form of aggression or hostile activity rather than an act of war, the line between this and warfare has become increasingly blurred in the age of generative AI. As discussed, both China and the US are concerned about AI’s ability to enhance their rival’s capacity to interfere in domestic politics. Neither would want to see their public opinion manipulated by the other.

To mitigate these risks, it would be helpful for the US and China to expand their channels of global dialogue on establishing rules and norms governing the use of generative AI by state actors – particularly in contexts involving foreign interference. A key component of this dialogue could be the clear definition and management of ‘cognitive warfare’ risks in the era of generative AI. A collaborative approach to establishing red lines for unacceptable state behaviour – while promoting a framework for the responsible, transparent, and ethical use of AI in international relations – would be a constructive step forward. Such cooperation could help reduce the risks of AI-fueled escalation and potentially contribute to the development of broader global standards, encouraging other nations to adopt safer and more responsible practices with generative AI technologies.

Footnotes

Acknowledgements

I would like to thank the guest editors of this special issue, Kai He and Huiyun Feng, for inviting me to contribute and for their helpful comments on this article. I am also grateful to the reviewers of both this paper and my earlier paper, “The US Factor in Chinese Perceptions of Militarized AI,” where I was encouraged to explore the concept of the security dilemma as a framing device. The manuscript went through several iterations in refining its theoretical focus and scope, and it would not have been possible without their insightful feedback and suggestions. Finally, for a paper examining the security threat of ChatGPT, it seems appropriate to acknowledge ChatGPT itself, whose polishing and proofreading functions were used to improve the grammar and editing of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Jinghan Zeng

Notes

References

Allison

(2015) The Thucydides trap: Are the U.S. and China headed for war? The Atlantic. Available at: http://www.theatlantic.com/international/archive/2015/09/united-states-china-war-thucydides-trap/406756/ (accessed 21 May 2016).

Beauchamp-Mustafaga

(2023) Chinese next-generation psychological warfare. RAND. Available at: https://www.rand.org/pubs/research_reports/RRA853-1.html (accessed 21 May 2025).

Beckley

Brands

(2022) Danger Zone: The Coming Conflict with China. New York: W. W. Norton & Company.

Bonfanti

(2022) Artificial intelligence and the offense–defense balance in cyber security. In: Cavelty

Wenger

(eds) Cyber Security Politics. London: Routledge.

Booth

Wheeler

(2008) The Security Dilemma: Fear, Cooperation and Trust in World Politics. Basingstoke: Palgrave Macmillan.

Booth

Mitzen

Schweller

(2011) Knowing the unknown unknowns: Misplaced certainty and the onset of war. Security Studies 20(1): 2–35.

Bradford

(2023) The race to regulate artificial intelligence: Why Europe has an edge over America and China. Foreign Affairs. Available at: https://www.foreignaffairs.com/united-states/race-regulate-artificial-intelligence-sam-altman-anu-bradford (accessed 9 November 2024).

Breuer

Johnston

(2019) Memes, narratives and the emergent US–China security dilemma. Cambridge Review of International Affairs 32(4): 429–455.

Butterfield

(1951) History and Human Relations. London: Collins.

10.

CAC (2023) Interim administrative measures for generative artificial intelligence services. Available at: https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm (accessed 28 October 2025).

11.

Chaoyang

(2023) 警惕美国利用人工智能加剧对华认知战 (Beware of the US using AI to intensify its cognitive warfare against China). The Paper. Available at: https://www.thepaper.cn/newsDetail_forward_21933225 (accessed 21 May 2025).

12.

Cheng

Zeng

(2023) Shaping AI’s future? China in global AI governance. Journal of Contemporary China 32(143): 794–810.

13.

China (2019) 人工智能具有很强的 ‘头雁’效应 (Artificial intelligence has a strong ‘head goose’ effect). Available at: http://paper.people.com.cn/rmrbhwb/html/2019-07/26/content_1938122.htm (accessed 28 Feburary 2020).

14.

China (2023) 起底美式人工智能丑闻 (Uncovering the American artificial intelligence scandal). China Daily. Available at: https://weibo.com/1663072851/4871144455013002 (accessed 28 October 2025).

15.

Christensen

(1999) China, the U.S.-Japan alliance, and the security dilemma in East Asia. International Security 23(4): 49–80.

16.

Christensen

(2006) Fostering stability or creating a monster? The rise of China and U.S. policy toward East Asia. International Security 31(1): 81–126.

17.

Copeland

(2000) The constructivist challenge to structural realism: A review essay. International Security 25(2): 187–212.

18.

De Luce

Grumbach

(2024) AI gives Russia, China new tools to sow division in the U.S., undermine America’s image, intel agencies say. NBC News. Available at: https://www.nbcnews.com/politics/national-security/china-russia-ai-divide-us-society-undermine-us-elections-power-rcna142880 (accessed 28 October 2025).

19.

Ding

(2018) Deciphering China’s AI dream. Available at: https://www.fhi.ox.ac.uk/deciphering-chinas-ai-dream/ (accessed 28 Feburary 2020).

20.

Ding

(2019) The interests behind China’s artificial intelligence dream. In: Wright

(ed.) Artificial Intelligence, China, Russia, and the Global Order: Technological, Political, Global, and Creative Perspectives. Maxwell AFB, AL: Air University Press, pp. 43–47.

21.

Friedberg

(2012) A Contest for Supremacy: China, America, and the Struggle for Mastery in Asia. New York: W. W. Norton & Company.

22.

Gao

(2023) ChatGPT的 ‘创造性破坏’效应及其风险应对 (ChatGPT’s ‘Creative Destruction’ Effect and Risk Response). 中国社会科学报. (Chinese Social Sciences Journal). Available at: https://www.cssn.cn/skgz/bwyc/202303/t20230306_5601321.shtml (accessed 9 November 2024).

23.

Garfinkel

Dafoe

(2019) Artificial intelligence, foresight, and the offense-defense balance. War on the Rocks. Available at: https://warontherocks.com/2019/12/artificial-intelligence-foresight-and-the-offense-defense-balance/ (accessed 21 May 2025).

24.

Glaser

(1997) The security dilemma revisited. World Politics 50(1): 171–201.

25.

Goldstein

Sastry

Musser

, et al. (2023) Generative language models and automated influence operations: Emerging threats and potential mitigations (A Joint Report with Georgetown University’s Center for Security and Emerging Technology OpenAI and Stanford Internet Observatory). Available at: https://cyber.fsi.stanford.edu/io/publication/generative-language-models-and-automated-influence-operations-emerging-threats-and (accessed 21 May 2025).

26.

GT (2024) 警惕！美军动向：与美科技巨头加速勾连，生成式AI军事化 (Beware! US military trends: Accelerating links with US tech giants, militarization of generative AI). Global Times. Available at: http://www.news.cn/mil/2024-01/30/c_1212331950.htm (accessed 28 October 2025).

27.

Herz

(1951) Political Realism and Political Idealism: A Study in Theories and Realities. Chicago, IL: University of Chicago Press.

28.

(2016) 军事指挥信息系统中的机器智能：现状与趋势 (Machine intelligence in military command information system: Current situation and trend). 学术前沿/Frontiers 8: 22–34.

29.

Huang

Yao

(2023) 被重塑的世界？ChatGPT崛起下人工智能与国家安全新特征 (Reshaping the world? AI and new features of national security under the rise of ChatGPT). 国际安全研究/Journal of International Security Studies 4: 82–106.

30.

Huang

Yao

(2024) 再塑造”与“高风险”生成式人工智能对舆论安全的影响 (The impact of ‘reshaping’ and ‘high-risk’ generative AI on public opinion security). 情报杂志/Intelligence Magazine 4.

31.

Jervis

(1976) Perception and Misperception in International Politics. Princeton, NJ: Princeton University Press.

32.

Jervis

(1978) Cooperation under the security dilemma. World Politics 30(2): 167–174.

33.

Yan

(2023) 人工智能的地缘政治：传播政治经济学的视角 (The geopolitics of AI: A political economy of communication perspective). 南昌大学学报/Journal of Nanchang University 6.

34.

Johnson

(2023) AI-security dilemma: Insecurity, mistrust, and misperception under the nuclear shadow. In: AI and the Bomb: Nuclear Strategy and Risk in the Digital Age. Oxford: Oxford University Press.

35.

Johnston

(2019) The failures of the ‘failure of engagement’ with China. The Washington Quarterly 42(2): 99–114.

36.

Kertzer

Brutger

Quek

(2024) Perspective taking and security dilemma thinking: Experimental evidence from China and the United States. World Politics 76(2): 334–378.

37.

Liu

(2023a) ChatGPT赋能数字法治政府建设中的风险及法律规制 (ChatGPT empowers the risks and legal regulations in the construction of digital rule of law government. 新兴权利/ Emerging Rights 2.

38.

Liu

(2023b) 生成式人工智能的三大安全风险及法律规制 (Three major security risks and legal regulations of generative artificial intelligence). 东方法学/Eastern Law 4: 29–43.

39.

Marcellino

Beauchamp-Mustafaga

Kerrigan

, et al. (2023) The rise of generative AI and the coming era of social media manipulation 3.0. RAND Report. Available at: https://www.rand.org/pubs/perspectives/PEA2679-1.html (accessed 21 May 2025).

40.

Mearsheimer

(2014a) The Tragedy of Great Power Politics (Updated Edition). New York: W. W. Norton & Company.

41.

Mearsheimer

(2014b) Can China rise peacefully? The National Interest. Available at: http://nationalinterest.org/commentary/can-china-rise-peacefully-10204 (accessed January 14 2015).

42.

NSCAI (2021) The National Security Commission on Artificial Intelligence (NSCAI) final report. Available at: https://reports.nscai.gov/final-report/ (accessed 8 November 2024).

43.

Xiang

(2023) 生成式人工智能——ChatGPT的变革影响、风险挑战及应对策略 (Generative artificial intelligence – ChatGPT’s impact, risks, challenges and countermeasures). 重庆大学学报社会科学版/Journal of Chongqing University Social Science Edition 3: 102–114.

44.

Qiao-Franco

Bode

(2023) Weaponised artificial intelligence and Chinese practices of human–machine interaction. The Chinese Journal of International Politics 16(1): 106–128.

45.

Shang

(2023) 生成式人工智能风险治理元规则研究 (Research on Meta-rules of generative artificial intelligence risk governance). 东方法学/Eastern Law 3: 4–17.

46.

Sheehan

(2018) How China’s massive AI plan actually works. Macro Polo. Available at: https://archivemacropolo.org/analysis/how-chinas-massive-ai-plan-actually-works/?rp=m (accessed 28 October 2025).

47.

Shepardson

(2023) US senate leader Schumer calls for AI rules as ChatGPT surges in popularity. Reuters. Available at: https://www.reuters.com/world/us/senate-leader-schumer-pushes-ai-regulatory-regime-after-china-action-2023-04-13/ (accessed 28 October 2025).

48.

SkyNews (2025) Donald Trump warns DeepSeek should be ‘wakeup call’ for America’s AI industry. Sky News. Available at: https://news.sky.com/story/donald-trump-warns-deepseek-should-be-wakeup-call-for-americas-ai-industry-13298001 (accessed 17 February 2025).

49.

Song

(2025) 人工智能：如何握好这把 ‘双刃剑’ (AI: How to use this double-edged sword). Guangming Daily. Available at: http://politics.people.com.cn/n1/2025/0417/c458474-40462064.html (accessed 21 May 2025).

50.

Steinberg

O’Hanlon

(2014) Strategic Reassurance and Resolve: U.S.-China Relations in the Twenty-First Century. Princeton, NJ: Princeton University Press.

51.

Wang

(2023) 生成式人工智能时代的政治传播走向——基于ChatGPT的讨论（The trend of political communication in the era of generative artificial intelligence: A discussion based on ChatGPT. 党政研究/Party and Government Research 3: 34–41.

52.

Suleyman

Bhaskar

(2023) The Coming Wave: AI, Power, and Our Future. New York: Random House Audio.

53.

Tang

(2009) The security dilemma: A conceptual analysis. Security Studies 18(3): 587–623.

54.

Tao

(2016) AlphaGo 技术原理分析及人工智能军事应用展望 (Principle analysis on AlphaGo and perspective in milltary application of artificial intelligence). 指挥与控制学报/Journal of Command and Control 2(2): 114–120.

55.

Waltz

(1979) Theory of International Politics. Reading, MA: Addison-Wesley.

56.

Wang

(2024a) 类Sora生成式AI对舆论生态的挑战及潜在风险 (The challenges and potential risks of Sora-like generative AI to the public opinion ecosystem). 法治网/Legal Network. Available at: https://www.cpd.com.cn/n15737398/n26490099/424/t_1132072.html (accessed 21 May 2025).

57.

Wang

(2024b) 人工智能诱发隐性意识形态话语风险的逻辑机理及化解策略 (The logical mechanism and mitigation strategies of the risks of implicit ideological discourse induced by AI). 马克思主义研究/Marxist Studies 4: 124–133.

58.

Webster

Creemers

Triolo

, et al. (2017) China’s plan to ‘lead’ in AI: Purpose, prospects, and problems. New America. Available at: https://digichina.stanford.edu/work/chinas-plan-to-lead-in-ai-purpose-prospects-and-problems/ (accessed 28 October 2025).

59.

Xie

Cao

(2024) 生成式人工智能意识形态安全风险探析 (An analysis of the ideological security risks of generative artificial intelligence). 国家安全研究/National Security Studies 1: 102–127.

60.

Zeng

(2020) Artificial intelligence and China’s authoritarian governance. International Affairs 96(6): 1441–1459.

61.

Zeng

(2021a) China’s artificial intelligence innovation: A top-down national command approach? Global Policy 12(3).

62.

Zeng

(2021b) Securitization of artificial intelligence in China. Chinese Journal of International Politics 14(3): 417–445.

63.

Zeng

(2022) Artificial Intelligence with Chinese Characteristics: National Strategy, Security and Authoritarian Governance. London: Palgrave Macmillan.

64.

Zeng

(2025) The US factor in Chinese perceptions of militarized AI. International Affairs 101(2): 677–689.

65.

Zeng

Mao

(2024) 认知武器化与人工智能认知战——一项机器学习与行为实验研究 (Cognitive weaponization and cognitive warfare of artificial intelligence: A study based on machine learning and behavioral experiment. 国际安全研究/Journal of International Security Studies 5: 49–80, 158, 159.

66.

Zhang

(2025) The promise and perils of China’s regulation of artificial intelligence. Columbia Journal of Transnational Law 63: 1–56.

67.

Zhu

Yang

(2023) ChatGPT/生成式人工智能与教育创新：机遇、挑战以及未来 (ChatGPT/AIGC and educational innovation: Opportunities, challenges, and the future). 华东师范大学学报教育科学版/Journal of East China Normal University Educational Sciences 7: 1–14.

68.

Zhuang

(2023) 涉嫌寻衅滋事罪！甘肃公安侦破首例利用AI人工智能技术炮制虚假信息案 (Suspected of provoking trouble! Gansu police cracked the first case of using AI technology to fabricate false information). 甘肃政法网/Gansu Politics and Law Network). Available at: http://www.chinapeace.gov.cn/chinapeace/c100064/2023-05/16/content_12657510.shtml (accessed 28 October 2025).