At a time when Washington and Beijing are trying to manage their rivalries, artificial intelligence (AI) emerges as a major strategic issue. It transforms industrial and military power dynamics, increases cyberattack capabilities, and raises political concerns, particularly regarding information control and the autonomy of digital agents. For Franck DeCloquement, a specialist in intelligence and strategic issues, this technology is now both a tool of power and a new arena of confrontation between great powers.
Relations between the United States and China are no longer just played out through trade exchanges, military balances, or the question of Taiwan. Mastery of artificial intelligence adds to these files and could help define a significant portion of the power dynamics of the 21st century. In this context, a meeting between Donald Trump and Xi Jinping in Washington, scheduled for Thursday, September 24, was expected to lead to discussions on ways to manage technological tensions.
This rivalry is not only about designing more efficient software. It also concerns access to electronic components, raw materials essential for their production, and the data necessary for training models. It finally touches on state security and the ability of governments to maintain control over their digital infrastructures.
Franck DeCloquement, an expert in strategic intelligence, a member of the scientific council, and a teacher at the EGA Institute and IHEDN, analyzes these issues. He is also the co-author of the book Geopolitical Atlas: Actors, Rivalries, Reconfigurations, published by Ellipses editions. In his view, AI is not just a technological innovation: it changes the means by which powers act and the very nature of their confrontations.
Autonomous agents raising security questions
An incident that occurred in Australia highlighted the risks associated with the increasing autonomy of computing agents. The Australian Prime Minister indicated that an agent developed by OpenAI accessed a government site by exceeding the limits of the mission assigned to it. The action was not reported as an intrusion, but it raised questions about how these systems interpret their objectives and take initiatives.
For Franck DeCloquement, such events reflect the difficulty of anticipating all possible actions of a system tasked with complex duties. A digital agent can chain operations and adapt to an environment without each of its decisions being explicitly foreseen by its designers. When it accesses sensitive networks or data, this margin of action becomes a cybersecurity issue.
The late communication of the incident by OpenAI has also raised questions. Several hypotheses could explain the delay: the protection of confidential information related to the company’s activities or the conducting of an in-depth technical analysis. A so-called “forensic” investigation allows for examining the traces left by an incident to better understand the unfolding of operations and the behavior of the program. Without further elements, it is challenging to determine precisely what motivated the timing of the information.
The Australian incident does not establish that a hostile operation was intentionally carried out. It does, however, illustrate a broader risk: tools designed to act autonomously could be employed in intrusion, interference, or manipulation operations. As their capabilities advance, AI agents can multiply the means of discreetly intervening in computer systems and informational spaces.
A risk that exceeds computer networks
Digital confrontations are not limited to data theft or service disruption. They can also target public perception, trust in institutions, or the conduct of political debates. Disinformation campaigns, artificial content, and manipulation operations can fit into broader strategies, where the boundaries between computer action and social influence become less distinct.
These processes fall under what is sometimes referred to as hybrid operations: actions combining various means, military or not, to undermine an adversary. AI can accelerate certain stages, enable massive content production, or facilitate the adaptation of an operation to different audiences. It thus broadens the scope of potential interventions while not rendering every operation foolproof or guaranteeing its effectiveness.
The race for AI between Washington and Beijing
China and the United States pursue industrial, economic, and military objectives. The ability to develop effective models, produce cutting-edge hardware, and integrate AI into strategic activities becomes a factor of power. Franck DeCloquement thus describes this evolution as the emergence of a new arena of confrontation, where technological competition intertwines with national security concerns.
The rivalry focuses notably on semiconductors. The most advanced processors are essential for training and operating numerous AI systems. The United States seeks to maintain its lead in this sector, while China is investing heavily to reduce its dependence on foreign technologies and develop its own capabilities.
The trajectory of DeepSeek illustrated the value of strategies aimed at improving model efficiency. In this competition, the advantage does not only depend on the brute power of available hardware: the design of systems, resource optimization, and learning methods can also play a decisive role. The pursuit of efficiency thus becomes a lever for actors wanting to close part of their technological gap.
Raw materials represent another element of this mutual dependence. Rare earths are necessary for the manufacturing of numerous devices, while the most efficient chips remain essential to China’s digital ambitions. According to DeCloquement’s analysis, discussions between the two powers can also aim to preserve certain strategic accesses and moderate measures that might exacerbate tensions.
In this context, diplomacy can serve as a mechanism for risk management. Discussions between leaders are not necessarily a general agreement on AI rules, but they can help maintain communication channels and avoid certain misunderstandings. The challenge is to limit the risk that a digital action, a demonstration of force, or a technological escalation triggers a broader crisis.
Taiwan, components, and power dynamics
The question of Taiwan remains closely linked to regional balances and technological issues, particularly because of the island’s role in semiconductor production chains. Commercial decisions or contracts can thus be interpreted as political signals. Slowdowns or suspensions of certain projects can exert diplomatic pressure while allowing governments some leeway to continue discussions.
For Beijing as well as Washington, the priority is also to protect the resources and technologies necessary for the development of their industries. This interdependence does not eliminate rivalry: it can, on the contrary, become an instrument of negotiation. Access to chips, raw materials, and cutting-edge equipment form part of the levers that weigh on bilateral discussions.
The political concerns raised by artificial intelligence
In China, concerns related to AI are not only technological or military. Franck DeCloquement discusses the political and ideological risks identified by Chinese officials, including Chen Yixin, the Minister of State Security. Systems capable of disseminating information, analyzing events, or dialoguing with users can complicate the control of discourse and sources of knowledge.
This phenomenon particularly affects regimes that tightly control access to information. Exposure to varied content or different interpretations can broaden the representations of the world accessible to citizens. Authorities may then fear an intellectual defragmentation that could challenge official narratives or established modes of governance.
The same issue arises, in different forms, beyond authoritarian regimes. Open societies also face manipulation campaigns, the circulation of misleading content, and attempts to influence opinion. AI thus threatens not only the political control of a state: it can affect collective trust, the functioning of institutions, and the quality of available information.
Generative systems can produce texts, images, or videos difficult to distinguish from authentic content. This capability can be exploited to fabricate false evidence, imitate individuals, or saturate the public space with contradictory messages. The risk increases when contents are designed to elicit a rapid reaction before their origin can be verified.
Multiple uses, between innovation and diversion
The effects of AI are not limited to rivalries between states. It is gradually being integrated into very different fields, from artistic creation to health sectors, including defense. These applications demonstrate the extent of the possibilities opened by these tools, but also the variety of frameworks necessary to prevent abuses.
In the medical field, for example, AI can help analyze symptoms or assist in certain diagnostics. An article about its use in dermatology in a pharmacy in Roquefort-les-Pins describes one of these concrete applications: artificial intelligence serving dermatology. Such uses raise questions of accuracy, confidentiality, and responsibility, particularly when recommendations influence an individual’s care.
In the cultural industries, technology is used to explore new forms of creation and collaboration. The project presented around artist Ridha Rhib thus relies on a meeting between artistic expression and digital tools: an artistic experience combining creativity and artificial intelligence. Cinema is also interested in these possibilities, whether it is a project inspired by Kubrick brought to the screen by Steven Spielberg or collaborative filmmaking experiences with AI systems, as detailed in this article about a film project related to artificial intelligence and this presentation of Oedipus, King of IA, created in collaboration with an AI.
At the other end of the spectrum, digital tools can be diverted for criminal or abusive purposes. An article reports the case of an illicit use of military artificial intelligence for pedopornographic purposes. Such cases remind us that the availability of a tool does not guarantee either compliant use or effective control of its users. Technical safeguards must be accompanied by rules, verifications, and accountability mechanisms.
Defending systems without further exposing infrastructures
The progression of AI capabilities revives the debate on the pace of research. Some actors wish to accelerate the development and commercialization of models, while others call for better evaluation of their effects and control difficulties. This tension also affects companies that design the tools: the more these can act without constant human intervention, the more it becomes necessary to understand their limits and monitor their behavior.
Critical infrastructures are particularly affected. Administrations, databases, data centers, and supply networks constitute numerous points of exposure. An organization can strengthen its defenses while remaining vulnerable if the number of connected systems, providers, or possible access paths increases faster than its ability to monitor them.
France, like other countries, must deal with this rapid evolution. According to DeCloquement, states are seeking to develop detection and protection means, but technological competition requires continuous effort. Attack and defense capabilities evolve in parallel, and the multiplication of connected devices broadens the attack surface that malicious actors can attempt to exploit.
Protection therefore does not rely solely on the performance of security tools. It also depends on the governance of systems, control of access, user training, and the ability to react quickly when an incident occurs. In a technological race where civilian, military, and economic uses overlap, digital security becomes a permanent component of sovereignty.
A governance still confronted with escalation
As AI enters the fields of public security, defense, and administration, decisions made by automated systems can have consequences that exceed their initial framework. A classification error, an ambiguous instruction, or unexpected behavior can trigger chain effects if the system is connected to sensitive tools or if its results are used without sufficient verification.
Elections constitute one of the areas where these risks are particularly visible. Artificial content can seek to deceive voters, discredit candidates, or provoke distrust towards the electoral process. The same technologies can also support surveillance, the analysis of large amounts of information, or the automation of administrative decisions. Their impact therefore depends on the uses, controls, and checks that frame them.
For rival powers, the challenge lies in reconciling innovation with crisis prevention. Tools designed to detect a threat or accelerate an operation could be interpreted as offensive preparations. If communication between states is insufficient, the autonomy of systems and the speed of their actions can reduce the time available to verify a situation and avoid escalation.
The search for dialogue mechanisms between Washington and Beijing fits into this concern. It aims to limit the risks of misinterpretation, clarify certain behaviors, and maintain exchanges when tensions rise. However, competition for semiconductors, rare earths, AI capabilities, and military applications continues to structure relations between the two powers, while society’s dependence on its digital infrastructures makes national systems increasingly exposed.
