September 2026
By Daryl G. Kimball

The danger of nuclear war continues to loom over humanity. Without new restraints, a more dangerous era lies ahead. The United States, China, and Russia are seeking to modernize and upgrade their arsenals to address perceived nuclear deterrence gaps, including by increasing deployed strategic arsenals and fielding “more usable” tactical nuclear weapons for regional war scenarios.

Into this volatile landscape comes artificial intelligence—not as a future concern, but as a present reality. Private technology companies and governments are caught up in a high-stakes competition to develop advanced frontier AI systems. Leading technologists warned in July that “they could be close to automating AI research,” which will accelerate AI development possibly “beyond our ability to understand or control the resulting systems.”

These two existential challenges—disarming AI and preventing nuclear war—can no longer be understood or governed in isolation.

As nuclear weapons competition and risk grow, the U.S., Chinese, and Russian militaries are racing to integrate AI tools into their conventional military systems and nuclear command-and-control systems to process more information more quickly to provide, as the Pentagon describes it, a “decision advantage” in a crisis or a war.

But the unconstrained use of AI tools may distort the information humans use and the choices they consider when making world-changing decisions regarding nuclear weapons. For example, AI-enabled warfare will compress the time available for meaningful human judgment. Commanders under pressure may be handed algorithm-generated menus of response options that neglect creative, human-centered solutions.

Leaders under stress are also vulnerable to automation bias—the tendency to trust a machine-generated information and analysis over one’s own judgment—even though AI systems are well known for projecting confidence when they are wrong. Another risk is that autonomous weapons with AI-enabled targeting plans could be used to attack nuclear weapons-related infrastructure and prompt a nuclear retaliation. India launched such an attack during its 2025 war with Pakistan.

These and other factors mean that the reckless integration of AI into military and nuclear operations will introduce new uncertainties to the already uncertain and unsustainable practice of nuclear deterrence. The likely introduction of recursive, self-improving, and agentic AI systems in the near future will only increase the risks of catastrophic outcomes.

Unfortunately, the political, diplomatic, and legislative responses to regulate these technologies, and to mitigate the nuclear risks they pose, are failing to keep pace. The best we have, for now, is the growing international consensus that machines must never supplant human judgment in the authorization or execution of nuclear weapons launches.

This “human in the loop” principle, while necessary, is insufficient. There are no agreed definitions, rules of the road, or accountability mechanisms. Nuclear-weapon states have, so far, failed to engage in sustained discussions on how they each plan to put the principle into practice and enforce it.

As a prestigious group of Nobel laureates, technologists, and nuclear experts recommended in an appeal issued from Rome in July, governments, with help from frontier AI developers, researchers, and the international community, need to act now to avoid the reckless integration of AI into nuclear operations.

There are steps that could be pursued to enhance mutual security, starting with a formal commitment by the major nuclear-armed states, through coordinated action or ideally a binding agreement, to retain effective human control over any decision to use nuclear weapons.

These states could also engage in technical talks on measures to insert automated “firebreaks” in advanced command-and-control systems to disallow AI-enabled actions leading to nuclear escalation. By taking concrete steps to move away from postures that call for prompt retaliatory nuclear counterattack and agreeing that the sole purpose of nuclear weapons is to deter nuclear attack, they could also reduce the risk of miscalculation.

Military officials from the nuclear-armed states should be required to report regularly on their use of AI-assisted tools in nuclear operations and the measures that they have implemented to mitigate risks. Independent audits should be conducted to ensure that best practices are being followed.

Progress toward adopting such nuclear risk reduction measures, as well as new and verifiable agreements to cap and reduce nuclear arsenals, is not possible without sustained, serious dialogue among the major nuclear-armed states. Since promising a “new era” of multilateral nuclear arms control after the New Strategic Arms Reduction Treaty expired in February, however, the Trump administration has failed to engage constructively with either China or Russia on nuclear weapons matters.

Although such dialogue may appear unlikely, it is necessary and possible. U.S. President Donald Trump and Chinese President Xi Jinping could set a new and safer course by launching high-level, bilateral talks on nuclear risk reduction and rules of the road for AI use in nuclear weapons systems at their summit this month in Washington.

Immediate and decisive action is required to counter these growing nuclear dangers before it is too late.