The Doomsday Clock &
Our Last Best Chance

Not all existential risks are equal in controllability. Some — like asteroid impacts — are largely a matter of detection and technology. Others — like nuclear war and pandemic preparedness — are almost entirely political and social choices. Understanding which levers we actually hold is the first step toward pulling them.

Nuclear war remains the threat that launched the Doomsday Clock in 1947 and still dominates it today. The 2026 Bulletin statement notes that multiple conflicts involving nuclear-armed states intensified in 2025 — including Russia-Ukraine, India-Pakistan, and US-Iran flashpoints. Crucially, the New START treaty — the last binding agreement limiting US and Russian nuclear arsenals — expired in February 2026, leaving no legal ceiling on warhead production for the first time in over half a century.

RAND Corporation researchers, analysing AI's role in nuclear scenarios, concluded that while AI could detonate every warhead in the 12,000-strong global stockpile, the resulting nuclear winter would likely fall short of outright extinction — humans are too dispersed. However, the humanitarian catastrophe would be civilisation-ending, and the cascading effects on food systems, disease, and climate could push much further.

This is the single threat most completely within human political control. Arms reduction treaties work. The 1991 Strategic Arms Reduction Treaty moved the Doomsday Clock from 3 minutes to 17 minutes overnight.

The Bulletin's 2026 statement documents a relentless worsening: atmospheric CO₂ reached 150% of pre-industrial levels, 2024 was the warmest year in 175 years of recorded history, and the hydrological cycle has become increasingly erratic — with catastrophic floods and droughts striking multiple continents simultaneously. For the third consecutive year, Europe recorded over 60,000 heat-related deaths.

Climate change differs from nuclear war in one critical respect: it is already happening, and its effects compound every other existential threat. It drives food and water scarcity, displaces populations (creating conditions for conflict), and may destabilise frozen permafrost containing ancient pathogens.

Of all existential threats, this is the one where AI demonstrates the clearest and most unambiguous benefit — from compressing 100-year climate projections from months to hours, to designing new low-carbon materials, to predicting extreme weather events with unprecedented precision.

Of all existential threats, AI-enabled biological risk is the most rapidly evolving and least governed. The Forecasting Research Institute reported in 2025 that biosecurity experts and superforecasters believe AI could make a global pandemic five times more likely if critical safety gaps are not closed. Researchers have already demonstrated that AI models can outperform PhD-level virologists on experimental virology questions.

The timeline for concern is short. Analysts forecast that by late 2026, AI systems may be capable of assisting domain experts in designing novel biological threats — and by 2028, providing meaningful support to non-expert bad actors. Anthropic has activated ASL-3 biological safety protocols, and OpenAI held a biodefense summit — but voluntary measures are lagging the pace of capability development.

The 2026 Bulletin statement specifically called for new multilateral agreements on biological threats — an international biosecurity framework equivalent in ambition to the nuclear non-proliferation treaty.

AI misalignment — the possibility that sufficiently advanced AI systems pursue goals misaligned with human values, potentially with catastrophic consequences — is the most contested existential risk. In 2024, hundreds of AI researchers signed a statement that "mitigating the risk of extinction from AI should be a global priority alongside nuclear war and pandemics."

The current Doomsday Clock statement specifically calls for international guidelines on the use of AI alongside nuclear and biological arms control. The 2026 Bulletin cited AI's integration into battlefield decision-making, social media manipulation at scale, and the accelerating pace of capability development as concrete near-term concerns — distinct from the longer-term alignment problem.

Unlike asteroid impacts or climate change, AI misalignment risk is genuinely self-referential: the tool we might use to solve it is the same tool creating the problem. Interpretability research — understanding what AI systems are actually doing inside — remains one of the most underfunded areas relative to capability development.

The UN Secretary-General has called for a legally binding ban on lethal autonomous weapons by 2026. The United States, notably, has rejected international oversight frameworks — a geopolitical fracture with direct existential implications.

Food and water scarcity are not independent threats — they are amplifiers of every other risk on this list. Climate change drives drought and agricultural collapse. Conflict interrupts supply chains. Population growth and mismanagement deplete aquifers. The 2026 Bulletin documented large-scale droughts across Peru, the Amazon, southern Africa, and northwest Africa, alongside record-breaking floods displacing hundreds of thousands.

These are also the threats where AI's humanitarian potential is clearest and least contested. TinyML sensors deployed in developing-world fields optimise irrigation in real time. IBM-Cleveland Clinic quantum-AI hybrid systems are being tested for fertiliser optimisation. Microsoft's MatterGen is specifically targeting materials that improve crop yields in harsh climates. Google's AI-enhanced weather forecasting provides critical seasonal outlook data for agricultural planning.

Unlike nuclear or AI misalignment risk, food and water solutions do not require solving deep geopolitical problems first. The technology exists; the challenge is deployment at scale, equitably distributed.

An asteroid impact of the scale that ended the dinosaurs is a low-probability but total-extinction event. NASA's DART mission in 2022 demonstrated that humanity can physically deflect a near-Earth object. The challenge is detection with sufficient advance warning — decades, ideally.

This is one area where AI provides unambiguous benefit with essentially no dual-use risk. AI dramatically accelerates the analysis of sky survey telescope data, trajectory modelling, and impact probability assessment. The Vera Rubin Observatory, coming online in 2025, will produce petabytes of sky survey data per night — unmanageable without AI filtering and classification.

Of all existential threats, this is the one humans are most technically equipped to solve — if political will and funding are maintained. The existential risk from asteroids today is primarily a risk of inattention, not technical inability.