Confronted with the vastly amplified consequences of algorithmic warfare, the unprecedented empowerment of non-state actors, and the deepening AI security divide, the only way forward is to shoulder responsibility, reject exclusive blocs, and pursue sincere cooperation.
by Liu Chong
Recently, the U.S. Department of Defense has announced an accelerated shift toward an “AI-prioritized” operational posture and signed cooperation agreements with eight leading U.S. tech companies, greatly accelerating the deployment of real-world intelligent systems — a move that has drawn international attention. Meanwhile, the risks of cyberattacks and threats to critical infrastructure exposed by the Mythos model have deepened global concerns.
Against the explosive advancement of artificial intelligence (AI) and the exponential rise of its risks, current international governance of military AI remains preoccupied with academic debates over “human-in-the-loop” mechanisms and the applicability of international law, moving at a leisurely pace with little meaningful consensus.
The world is left to ask: What truly matters for AI security governance? Can existing international norms address these challenges?
Confronted with the vastly amplified consequences of algorithmic warfare, the unprecedented empowerment of non-state actors, and the deepening AI security divide, the only way forward is to shoulder responsibility, reject exclusive blocs, and pursue sincere cooperation. This will ensure AI advances for the betterment of human civilization. China’s core principles — putting people first, using AI for good purposes, prioritizing development, and ethics first — carry ever greater significance.
DOES “HUMAN-IN-THE-LOOP” SOLVE AI SECURITY RISKS?
Since 2014, when international discussions on lethal autonomous weapons systems began under the United Nations Convention on Certain Conventional Weapons, questions such as whether fully autonomous lethal robots should be permitted and whether humans must remain in the decision loop have become central to global military AI governance.
Broadly, these debates boil down to whether existing international law applies to AI security. The U.S. Department of Defense has promoted a “responsible AI” strategy centered on establishing a complete accountability chain for military AI. Its core logic emphasizes traceability: whether humans remain in the decision loop, whether clear accountable actors exist, and whether post-incident accountability is possible.
Yet in an era where AI revolutionizes the speed, scale, and complexity of warfare, focusing solely on “human-in-the-loop” increasingly fails to capture the true humanitarian risks of intelligentized conflict — and risks normalizing severe humanitarian consequences. Accountability traceability does not equal acceptable outcomes.
What truly defines modern warfare is not whether machines fire autonomously, but how AI transforms the scale, tempo, and escalation dynamics of conflict in unprecedented ways. More critical than human presence in the loop is establishing new international norms to constrain the devastating consequences of algorithmic warfare.
Even when humans retain final approval, AI systems can generate hundreds or thousands of potential targets in seconds and recommend operational plans in real time across the “sense-decide-strike” chain. AI does not merely replace humans; it exponentially amplifies human capacity for violence. Think tanks warn that this pattern of “rapid recommendations, cursory reviews” reduces human judgment to a rubber stamp. Much like U.S. drone operators once likened strikes to video games, today’s commanders risk becoming desensitized to the humanitarian costs of their orders.
Israel’s deployment of AI-driven algorithms to conduct indiscriminate attacks in Gaza, resulting in unprecedented humanitarian catastrophe, offers a stark lesson. The United States and Israel, leveraging AI intelligence capabilities, carried out the targeted killing of Iran’s top leader, dragging Iran and Gulf nations into conflict and triggering a global energy crisis. These cases prove that the gravest danger of algorithmic warfare lies not in full autonomy, but in a vastly accelerated escalation and vastly expanded strike scale.
Legally, traditional international laws governing warfare were crafted under the assumption of limited human decision-making capacity. The AI era has fundamentally altered this premise. Future governance must move beyond “who is accountable” to design new international norms that compel decision-makers to exercise restraint when wielding hyper-empowered military tools.
This cannot be achieved by any single country or discipline alone. It requires great powers to assume responsibility, and international organizations to coordinate action, and calls for interdisciplinary collaboration across international law, AI, cybersecurity, strategic studies, and public policy. At its core, this is a global governance project rooted in putting people first and prioritizing ethics.
ADDRESSING THREATS FROM NON-STATE ACTORS
Beyond state-to-state conflict, AI’s massive empowerment of non-state actors has emerged as a pressing concern. Low-cost drones, advanced AI models, automated cyberattacks, and target recognition capabilities — tools once exclusive to states — are becoming cheaper and widely accessible, also to civilians.
The recent release of the Claude Mythos model has sparked global alarm. The U.S. Treasury and regulatory bodies fear that such advanced models could enable large-scale cyberattacks against financial systems and critical infrastructure.
Multiple international think tank reports repeatedly warn that cutting-edge AI amplifies risks across cyberattacks, biological threats, information manipulation, and autonomous weapons. AI not only reshapes military balances between states but also lowers barriers for non-state actors to inflict mass harm.
Preventing malicious use of low-cost intelligent platforms has become an urgent global governance imperative. U.S. Treasury Secretary Scott Bessent has noted that China and the United States are discussing AI safety guardrails, seeking best practices to block access to advanced models by terrorist groups, hackers, and other non-state actors.
“AI for good” is no empty slogan. AI itself is neither inherently good nor evil; its impact depends entirely on human use. Just as AI fuels a surge of “one-person enterprises” driving economic prosperity, malicious actors could wield AI to unleash unprecedented destruction. Only through genuine international cooperation and the establishment of effective, universally accepted norms and safety guardrails can the world mitigate AI-driven threats from non-state actors.
BRIDGING THE AI DIVIDE THROUGH JOINT EFFORTS
AI represents a new frontier for humanity. The uneven distribution of AI capabilities constitutes the greatest inequity and strategic risk of the AI era for small and medium-sized countries.
Many nations lack the capacity to deter algorithmic warfare threats from major powers or defend against AI-enabled non-state actor attacks, leaving them unable to build robust cybersecurity or critical infrastructure defenses. NVIDIA CEO Jensen Huang recently argued that defending against powerful AI systems like Mythos requires coordinated action by open-source AI agents to safeguard cyberspace. His vision of leveraging open-source models and agent ecosystems to strengthen global cybersecurity offers valuable insights.
History shows that technology export controls are never foolproof. As AI-powered cyberattacks, automated vulnerability exploitation, and critical infrastructure attacks grow more sophisticated, restricting technology diffusion alone will prove ineffective. The real danger lies not in technology spreading, but in severe imbalances in defensive capabilities. What truly matters is ensuring all nations can defend against AI risks.
If only a handful of major countries possess advanced AI security capabilities while most countries lack basic defenses, the global security architecture will grow increasingly fragile. Building global cybersecurity capacity, promoting open cooperation on AI security tools, strengthening defensive systems in small and medium-sized countries, and establishing international AI risk early warning and coordination mechanisms must become priorities for global governance.
History has repeatedly shown that problems arising from development can only be solved through development. Exclusionary policies, unilateral dominance, and closed blocs only erode collective security. Prioritizing development and pursuing sincere cooperation are the only way to safeguard national security in the AI era.
In an age of carbon-silicon transformation, when hegemonic rivalries persist, no nation can alone address the systemic risks posed by advanced AI. The world’s greatest challenge is not AI competition, but preventing escalation and safeguarding shared security amid competition. Only by embracing a community with a shared future for humanity, standing united, and supporting one another can humanity navigate the turbulent waves of the AI era and steer civilization forward.
Editor’s note: Liu Chong is a research fellow and director of the Center for Science and Strategic Studies at the China Institute of Contemporary International Relations.
The views expressed in this article are those of the author and do not necessarily reflect those of Xinhua News Agency.
