The Two Horsemen of 2100: Why Experts Rank AI Misuse and Nuclear War as Humanity’s Greatest Threats

Humanity has always lived with the quiet awareness that its story could end, yet the terms of that ending have shifted dramatically across the centuries. Where plague, famine, and the sword once dictated the boundaries of survival, the twenty-first century introduces adversaries of our own making, engineered with brilliance and wielded with terrifying indifference.

A coalition of international experts now warns that by 2100, two forces stand above all others as existential threats: the deliberate misuse of artificial intelligence and the eruption of nuclear conflict. This is not idle speculation from fringe voices; it is a sober assessment emerging from serious minds who study catastrophe for a living.

The warning carries a peculiar weight because it names two dangers that are simultaneously technological and deeply human. Nuclear weapons are old enough to feel familiar, yet their capacity for annihilation has never diminished, only multiplied across unstable regions and fragile command systems. Artificial intelligence, by contrast, is young, accelerating, and poorly understood even by those who build it.

Together they form a pincer of risk, one ancient and one newborn, both capable of ending civilization within hours rather than decades. The convergence deserves far more attention than it currently receives in public discourse.

What makes this warning genuinely unsettling is its restraint. The experts did not offer a lurid doomsday narrative or a precise countdown; they simply identified the two threats most likely to extinguish humanity before the century closes. That minimalism is itself a message. When serious analysts strip away the noise and name only two dangers, the implication is that everything else, however alarming, remains secondary. Understanding why these two risks dominate the extinction ledger is the first step toward confronting them with the urgency they demand.

TL;DR International experts now rank the deliberate misuse of artificial intelligence and nuclear war as the two greatest threats to human extinction by 2100. Nuclear arsenals remain dangerously large and politically volatile, while AI advances faster than the safety frameworks meant to contain it. Both risks share a common thread: catastrophic outcomes depend on human decisions made under pressure, secrecy, and incomplete information. Mitigation requires stronger international governance, transparent AI development, and renewed arms control. The warning is not a prophecy but a prioritization, urging societies to treat these two dangers as the defining security challenges of the coming decades.

The Anatomy of an Extinction Warning

When a group of international experts convenes to rank the threats most likely to end human civilization, the resulting list functions as a kind of strategic compass. It tells governments, researchers, and citizens where to direct attention and resources. The 2100 horizon is deliberate, far enough to allow meaningful action yet close enough to demand it now.

Naming AI misuse and nuclear conflict as the top two risks is not a rhetorical flourish but a calculated judgment about probability and consequence.

Why Experts Converge on Two Threats

Extinction-level risks share a defining characteristic: they are global, fast-moving, and largely irreversible. A pandemic can be survived, an economic collapse can be rebuilt, and even severe climate disruption unfolds over generations. Nuclear war and runaway AI, however, can compress catastrophe into hours or days, leaving no room for recovery. That temporal compression is precisely why experts elevate them above slower-moving dangers that dominate everyday headlines.

There is also a structural reason for the convergence. Both threats are anthropogenic, meaning they originate in human ingenuity rather than natural forces. This makes them uniquely controllable in theory and uniquely dangerous in practice, because control depends on institutions that are fragmented, competitive, and often short-sighted. The experts are effectively warning that our greatest creations may outpace our greatest safeguards.

The Limits of the Warning

The original statement offered no detailed methodology, no probability estimates, and no ranked list beyond the top two. That absence of granularity is worth noting, because it reflects the genuine uncertainty surrounding existential risk assessment. Researchers can identify plausible pathways to catastrophe without being able to assign precise odds. The warning is therefore best read as a prioritization framework rather than a forecast with confidence intervals.

This restraint should not be mistaken for vagueness. In the field of existential risk, identifying the dominant threats is itself a major contribution, because it shapes where funding, regulation, and diplomatic energy flow. A warning that names two dangers clearly is more actionable than a sprawling catalogue of every conceivable apocalypse, and it forces institutions to confront uncomfortable trade-offs between competing priorities.

Artificial Intelligence as a Weapon and an Accident

Artificial intelligence occupies a strange position in the existential risk landscape because it is simultaneously a tool of immense promise and a potential instrument of annihilation. The experts specifically flagged deliberate misuse, a phrase that encompasses everything from autonomous weapons deployed without human oversight to AI systems repurposed for cyber warfare, disinformation, or bioweapon design.

The danger is not that machines will spontaneously decide to destroy us, but that humans will use them to do so with unprecedented efficiency.

Deliberate Misuse Versus Accidental Harm

The distinction between intentional misuse and accidental harm matters enormously for policy. Deliberate misuse implies an actor, a motive, and a target, which makes it addressable through deterrence, attribution, and international law. Accidental harm, by contrast, emerges from complexity, misalignment, or unforeseen interactions between systems, and it resists traditional security frameworks because there is no adversary to deter. Both pathways lead toward catastrophe, but they demand fundamentally different responses.

Consider an AI system integrated into military command structures. If an adversary manipulates its inputs, the result is deliberate misuse. If the system misinterprets ambiguous data and recommends escalation, the result is accidental harm. In practice, the two blur together, because adversaries may deliberately exploit the very ambiguities that cause accidents. This entanglement is what makes AI risk so difficult to govern and so urgent to study.

The Governance Gap

Regulatory frameworks for artificial intelligence remain fragmented, voluntary, and largely national, while the technology itself is global and increasingly commoditized. A model trained in one jurisdiction can be deployed in another, fine-tuned in a third, and weaponized in a fourth, all within weeks.

This asymmetry between technological velocity and institutional inertia is the core governance gap that experts fear most, because it leaves the world perpetually reactive rather than proactive.

Efforts such as the European Union's AI Act and various United Nations dialogues represent genuine progress, yet they cover only a fraction of the risk surface. Autonomous weapons, frontier model safety, and compute governance remain contested terrain. Without binding international agreements, the incentive structure rewards speed over caution, and the actors most willing to cut corners gain temporary advantage at the expense of collective security.

Nuclear Conflict in a Multipolar World

Nuclear weapons have haunted humanity since 1945, yet the risk of their use has never been static. The Cold War produced a bipolar standoff governed by mutual assured destruction, a grim but stable logic that kept the peace through terror. Today's landscape is multipolar, with established powers modernizing arsenals, regional states expanding capabilities, and non-state actors perpetually seeking access. The stabilizing assumptions of the past no longer hold, and the margin for error has narrowed considerably.

The Erosion of Arms Control

Decades of arms control treaties once provided transparency, verification, and predictable limits on arsenals. Many of those agreements have expired, been abandoned, or rendered obsolete by technological change. Without them, nuclear powers operate with less mutual visibility, which breeds suspicion and accelerates worst-case planning.

The disappearance of these guardrails is not merely a diplomatic inconvenience; it removes the very mechanisms that historically prevented miscalculation from becoming catastrophe.

Hypersonic delivery systems, cyber vulnerabilities in command-and-control networks, and the integration of AI into early-warning systems compound the danger. Decision windows that once spanned hours now shrink to minutes, leaving leaders with less time to verify information and more incentive to act preemptively. Each technological advance that compresses response time increases the probability that a false alarm or a misinterpreted signal triggers an irreversible response.

Regional Flashpoints and Escalation Ladders

Nuclear risk is not evenly distributed across the globe. Specific regional rivalries carry elevated potential for escalation, where territorial disputes, historical grievances, and fragile command structures intersect. In such theaters, a conventional skirmish can escalate rapidly through a ladder of responses, each step appearing rational to the actor taking it while collectively propelling the situation toward the nuclear threshold.

The danger is compounded by the fact that escalation ladders are rarely visible to outsiders. Leaders operate under time pressure, incomplete intelligence, and domestic political constraints that reward toughness over restraint. The result is a system in which catastrophic outcomes can emerge not from malice but from a cascade of individually defensible decisions, each one narrowing the space for de-escalation until none remains.

Where AI and Nuclear Risk Intersect

The two threats do not exist in separate silos; they interact in ways that amplify both. Artificial intelligence is already embedded in early-warning systems, intelligence analysis, and military planning, which means an AI failure could directly trigger a nuclear response. Conversely, nuclear brinkmanship creates the kind of high-pressure, time-compressed environment in which AI errors become most consequential. The intersection is where the most dangerous scenarios live.

Automation and the Speed of Decision

When AI systems accelerate the observe-orient-decide-act loop, they compress the time available for human judgment. A machine that detects an incoming attack and recommends retaliation within seconds leaves leaders with little room to question the recommendation. The danger is not that AI will launch weapons on its own, but that it will narrow the window in which humans can prevent a catastrophic mistake.

This dynamic creates a paradox: the same speed that offers military advantage also erodes the safeguards that prevent accidental war. Slower, more deliberate decision-making is often dismissed as weakness, yet it is precisely what allows verification, communication, and de-escalation. Preserving human deliberation in nuclear command is not technological nostalgia; it is a safety requirement with existential stakes.

Shared Vulnerabilities, Shared Solutions

Both AI and nuclear risk share a common vulnerability: dependence on complex systems that can fail in unexpected ways. Both also share a common solution: transparency, verification, and international cooperation. The same norms that make nuclear arsenals safer, such as mutual inspection and crisis communication, can be adapted to AI governance, provided the political will exists to build them.

Framing the two threats together also creates opportunities for coalition-building. States that might resist AI regulation for competitive reasons may accept it when framed as nuclear risk reduction, and vice versa. The intersection is not merely a source of danger; it is a potential lever for diplomacy, allowing progress on one front to reinforce progress on the other.

Governance, Deterrence, and the Human Factor

Every existential risk ultimately passes through human hands, which means governance and psychology matter as much as technology. Deterrence theory assumes rational actors who weigh costs and benefits, yet real leaders operate under stress, ideology, and domestic pressure. The gap between theoretical rationality and actual behavior is where catastrophic miscalculation thrives, and it is precisely the gap that institutional design must address.

Building Institutions That Outlast Crises

Effective governance for existential risk must survive changes in leadership, shifts in public mood, and the inevitable crises that distract attention. Treaties, verification regimes, and hotlines work because they embed restraint in structures rather than relying on individual virtue. The challenge is building such structures for AI, a domain that evolves faster than any treaty cycle can accommodate.

Adaptive governance offers one path forward, combining binding principles with flexible implementation. Rather than attempting to regulate specific technologies that may soon be obsolete, frameworks can target capabilities, behaviors, and outcomes. This approach preserves room for innovation while maintaining guardrails, and it allows international bodies to respond to new developments without reopening every negotiation from scratch.

The Psychology of Existential Risk

Human beings are notoriously poor at responding to threats that are abstract, distant, or probabilistic. Nuclear war and AI catastrophe both fit this profile, which helps explain why they receive less attention than immediate crises. Overcoming this psychological bias requires narrative, leadership, and institutional pressure that make the abstract feel urgent without descending into fatalism or paralysis.

There is also a deeper philosophical dimension. Accepting that our own creations could end us forces a reckoning with the limits of foresight and the responsibilities of stewardship. The experts' warning is ultimately an invitation to maturity, a call to govern our most powerful technologies with the seriousness they demand rather than the complacency they currently receive.

Toward a Realistic Mitigation Agenda

Warnings without action are merely elegies, so the practical question is what can be done. Mitigation requires simultaneous progress on multiple fronts: arms control, AI safety research, international institutions, and public engagement. No single intervention eliminates the risk, but a portfolio of measures can reduce the probability of catastrophe from near-certain to merely possible, which is the difference between survival and extinction.

Priorities for the Coming Decade

The most urgent priority is preserving and rebuilding arms control architecture before further erosion makes it unrecoverable. Parallel to that, AI safety research must shift from a niche academic pursuit to a mainstream engineering discipline with dedicated funding, standards, and career pathways. Both efforts require political leadership willing to invest in prevention rather than reaction.

International coordination is the connective tissue that binds these priorities together. Without shared norms, verification mechanisms, and crisis communication channels, national efforts will remain fragmented and easily circumvented. The challenge is not technical feasibility but political will, and generating that will requires sustained public pressure and informed leadership.

The Role of Individuals and Institutions

Individuals are not powerless in the face of existential risk. Citizens can demand accountability from leaders, support organizations working on these issues, and resist the fatalism that treats catastrophe as inevitable. Institutions, meanwhile, can embed risk awareness into their cultures, ensuring that decisions with existential consequences receive the scrutiny they deserve rather than being delegated to inertia.

The experts' warning is ultimately a statement about agency. By naming AI misuse and nuclear conflict as the top threats, they are also implying that these threats are addressable, that human choices still matter, and that the future remains open. Whether that openness leads to survival or ruin depends on whether the warning is heard in time and acted upon with the seriousness it demands.