Amna Saqib
The scheduled meeting between US President Donald Trump, Chinese President Xi Jinping and Russian President Vladimir Putin on the sidelines of the Asia-Pacific Economic Cooperation (APEC) summit in Shenzhen comes at a difficult moment for global nuclear arms control. With the expiration of the New Strategic Arms Reduction Treaty (New START) in February 2026, the US and Russia no longer operate under binding limits on their strategic nuclear arsenals.
The interest of Washington in bringing China into a new arrangement reflects a changing distribution of nuclear capabilities, but extending the existing bilateral framework into a trilateral one will not, by itself, address the wider conditions shaping strategic stability. The challenge is not simply how to limit three nuclear arsenals, but how to account for the technological developments, threat perceptions and regional security relationships that influence how states assess their deterrents.
The three powers approach the question from fundamentally different strategic positions. US wants China included in future arms-control arrangements, arguing that any framework which excludes a growing nuclear arsenal would be incomplete. Meanwhile, Beijing, maintains that its stockpile remains considerably smaller than those of the US and Russia and that the two largest nuclear powers should undertake further reductions first.
Its position reflects the principle that nuclear disarmament should take account of the existing imbalance and the security concerns of all parties, as outlined in Chinese working paper on the implementation of the Nuclear Non-Proliferation Treaty. Moscow, for its part, has an interest in preserving strategic arms limitations but also views their future through the wider deterioration of its relationship with Washington and the changing military balance.
These differences are not merely diplomatic disagreements. They reflect distinct assessments of what constitutes a credible and survivable deterrent. A framework based on numerical parity between three powers would therefore be difficult to construct, since each measures the adequacy of its forces against more than one strategic counterpart.
The difficulty extends beyond the number of warheads. The nuclear deterrence increasingly depends on whether states believe their retaliatory capabilities would survive an attack and remain effective. Moreover, missile defence can raise doubts about the ability of a smaller arsenal to penetrate an opponent’s defences, potentially encouraging further investment in offensive capabilities.
Precision-strike systems, meanwhile, create concerns about the vulnerability of strategic assets, particularly when conventional weapons could threaten facilities associated with nuclear command and control. Cyber operations and space-based surveillance introduce additional uncertainties by potentially affecting early-warning systems, communications and the information available to decisionmakers.
These developments do not have identical effects, but together they complicate the assumptions on which traditional arms-control arrangements were built. An agreement that restricts warhead numbers while overlooking these changing military capabilities may limit arsenals without addressing the reasons states continue to expand or modernise them.
Moreover, Artificial intelligence adds another layer to this problem. Its growing integration into intelligence assessment, early warning, surveillance and military decision-support is changing how armed forces process information and interpret potential threats. The principal concern is not that machines will independently decide to launch nuclear weapons, but that AI-assisted assessments could influence human judgement during a crisis.
Unreliable data, algorithmic errors or excessive confidence in automated outputs may distort the interpretation of an ambiguous warning, particularly when decisionmakers have limited time to assess the situation. Faster information processing does not necessarily produce better decisions when the consequences of error are severe. Research by SIPRI on the impact of military AI on nuclear escalation risk highlights how even non-nuclear applications of military AI can compress decision-making timelines and introduce additional risks.
Managing these challenges will require more than general commitments on responsible AI. It will depend on maintaining meaningful human oversight, reliable communication between nuclear-armed states and safeguards against technological interference with nuclear command-and-control systems.
This does not mean that comprehensive regulation of military AI should become a precondition for nuclear negotiations. Such an undertaking would be difficult to verify, particularly given the dual-use nature of many AI applications. A more practical starting point would be to address how these technologies are incorporated into nuclear decision-making.
The major powers could explore understandings on maintaining human responsibility for nuclear-use decisions, establishing communication channels for significant technological incidents and discussing safeguards for AI-assisted early-warning systems. Greater transparency need not involve disclosing sensitive operational details.
Even limited exchanges on risk-reduction practices could help reduce the possibility that technological failures or ambiguous signals are mistaken for deliberate action. Such measures would not resolve the underlying strategic competition, but they could make its management less vulnerable to miscalculation.
There is also a wider question of who is affected by an agreement negotiated among the three largest nuclear powers. Their decisions do not remain confined to their own strategic relationship. The Chinese force posture influences Indian assessment of its security environment, while India’s capabilities and doctrinal choices shape deterrence calculations of Pakistan.
A trilateral arrangement could therefore alter the strategic circumstances of states that have no direct role in negotiating it. This matters particularly in South Asia, where short missile flight times, dual-capable delivery systems and increasingly sophisticated conventional capabilities complicate the distinction between conventional and nuclear escalation.
The May 2025 India-Pakistan crisis illustrated how quickly conventional military operations involving missiles and drones can intensify a confrontation between nuclear-armed states. An approach to global arms control that overlooks these interconnected security relationships risks creating new uncertainties even as it seeks to reduce competition among the major powers. This does not mean that other nuclear-armed states should determine the terms of a US-China-Russia agreement, but their legitimate security concerns cannot be treated as irrelevant to the wider objective of strategic stability.
The statement of United Nations Secretary General following expiration of New START rightly emphasised the need to restore verifiable limits and prevent an unconstrained nuclear competition. That objective remains important, but numerical limitations alone cannot address the changing character of deterrence.
So, this Shenzhen meeting is less to be evaluated on the basis that that the three leaders sign up on an agreement and more by whether they establish a basis for sustained engagement. A realistic first step would be to reaffirm the importance of strategic arms control, initiate working-level discussions on nuclear risk reduction and explore practical understandings on early warning, command and control and emerging technologies.
These efforts should proceed alongside discussions on future numerical limitations rather than replace them. Ultimately, strategic stability cannot be secured by negotiating the size of nuclear arsenals in isolation from the technological and geopolitical conditions that shape their use. The more useful outcome from Shenzhen would be a recognition that managing nuclear competition requires a broader conversation than three powers and their warhead numbers alone.




