A global panel of experts has identified the malicious use of artificial intelligence and nuclear conflict as the two greatest threats to human survival by the year 2100, warning that without urgent governance both risks could converge into catastrophe within decades.

The report, compiled by researchers studying existential and catastrophic risks, singles out AI misuse not as a distant science-fiction scenario but as a near-term danger already taking shape: autonomous weapons systems, AI-designed bioweapons, disinformation capable of destabilizing governments, and the erosion of human oversight over critical military and infrastructure decisions. Nuclear conflict, meanwhile, remains the most familiar and best-studied existential threat, one that has haunted policymakers since the mid-twentieth century but which experts say has been dangerously neglected as global attention has shifted elsewhere.

A Threat That Never Disappeared

Nuclear weapons have threatened human extinction since the Cold War, when the United States and the Soviet Union amassed arsenals capable of destroying civilization many times over. By the early 1970s, the Soviet Union had overtaken the United States in intercontinental ballistic missiles and submarine-launched ballistic missiles, prompting an escalating cycle of innovation: multiple independently targetable re-entry vehicles capable of striking several targets with a single missile, followed by Soviet SS-20s and American cruise missiles designed to slip under enemy radar.

That arms race gave rise to mass protest movements, including Britain's Campaign for Nuclear Disarmament, founded in 1958, which pushed for unilateral disarmament through rallies and the annual Aldermaston march. Governments resisted, arguing that abandoning weapons unilaterally would leave their nations exposed, and insisted that any disarmament must be negotiated multilaterally among all major powers.

That fundamental standoff, the tension between the logic of deterrence and the catastrophic risk of miscalculation, remains unresolved. Experts behind the new report argue that the world has grown complacent, treating nuclear risk as a solved problem of a bygone era rather than an active and evolving danger, particularly as more states acquire weapons and as arms control treaties lapse or go unrenewed.

The New Variable: Artificial Intelligence

What distinguishes this report from earlier warnings is the pairing of nuclear risk with AI misuse as a comparably severe threat. Unlike nuclear weapons, whose destructive mechanics are well understood even if their political management is fraught, AI presents a moving target. Its capabilities are evolving faster than regulatory frameworks can adapt, and its risks are diffuse: a single malicious actor, rather than a state with a missile arsenal, could in theory use AI tools to design pathogens, manipulate financial systems, or compromise the command-and-control infrastructure that underpins nuclear deterrence itself.

That last possibility, the entanglement of AI with nuclear systems, is especially alarming to experts. Early warning systems and decision-support tools increasingly rely on machine learning, raising the prospect that an algorithmic error or an adversarial manipulation could trigger a nuclear response based on false information, compressing the time available for human judgment in a crisis.

Lessons From the Climate Debate

The report's call for new global governance echoes a pattern seen in past environmental diplomacy. When climate change first emerged as a scientific concern, industrialists and skeptical politicians dismissed it as natural variation, delaying action for years even as evidence mounted. It took the 1992 Earth Summit in Rio de Janeiro, attended by representatives of 178 nations, and the 1997 Kyoto Protocol to establish binding international targets. Even then, enforcement proved difficult: the United States Senate rejected ratification of the Kyoto Protocol, and many signatories struggled to meet their commitments.

Experts who study catastrophic risk warn that AI and nuclear governance could follow the same trajectory: broad recognition of danger, followed by fragmented, voluntary, and ultimately insufficient international agreements. Some researchers have floated proposals analogous to the adaptation funds and enforcement courts once suggested for climate change, including international bodies empowered to monitor AI development and penalize states or companies that cross agreed safety thresholds.

What Comes Next

The report does not suggest these risks are inevitable. Rather, it frames the coming decades as a narrow window in which coordinated governance, treaty renewal, and technical safeguards could meaningfully reduce the odds of catastrophe. That requires, experts say, treating AI safety with the same seriousness that nuclear arms control eventually received, rather than waiting for a crisis to force the issue.

The challenge, as with climate change and nuclear disarmament before it, lies less in diagnosing the danger than in persuading governments, corporations, and publics to act on warnings before the risks become irreversible.