Lancet Assessment: Malicious AI and Nuclear War Could End Humanity by 2100

A new global risk assessment published under the Lancet banner concludes that humans could be extinct by the year 2100 through malicious use of AI or nuclear conflict. The two hazards top a list of 17 major threats to human health over the next 75 years.

The assessment comes from a newly established commission, which identifies 17 catastrophic threats in total. Beyond malicious AI and nuclear war, the list includes cuts to aid spending by wealthy countries, the climate crisis and childhood malnutrition. The commission pairs the warnings with a procedural claim that carries equal weight: it says a “realistic route to act” on each of the 17 threats exists.

That sentence — that every threat on the list has a workable response — is the analytical hinge of the document. It reframes the assessment from an exercise in doom cataloguing into a policy agenda with a deadline measured in decades, not centuries.

Why does a medical journal assess extinction risk?

The Lancet occupies an unusual position among scientific publishers. Over the past two decades it has run standing commissions on topics that sit at the intersection of health, economics and geopolitics, including climate change and pandemic preparedness. Extinction-level risk enters its remit through the definition the commission uses: threats to human health at a civilizational scale.

Under that definition, a nuclear exchange is a health catastrophe before it is a security one. Malicious use of AI belongs on the same list because the commission treats engineered harm — whether through autonomous weapons, engineered pathogens enabled by AI tools, or infrastructure disruption — as a public-health endpoint, not merely a technical failure mode.

The 75-year horizon matters. It spans the working lifetime of children born this year, which is the framing risk bodies increasingly use to convert abstract probability estimates into governance timelines.

What are the 17 threats?

The commission identifies 17 major threats to human health through 2100. The assessment names several explicitly:

• Malicious use of AI — ranked at the top of the list
• Nuclear conflict — ranked alongside it
• Cuts to aid spending by wealthy countries
• The climate crisis
• Childhood malnutrition

The placement of aid cuts and malnutrition on the same list as nuclear war signals the commission's methodology. It scores threats by their aggregate effect on human health, not by their dramatic value. Malnutrition already kills; reduced aid spending compounds that toll within years, not decades.

The prominence of malicious AI reflects how far the governance conversation has moved. A medical commission now ranks deliberate misuse of artificial intelligence as a co-equal extinction-class risk with nuclear weapons — a framing that would not have appeared in a health-sector document a decade ago.

What does the assessment actually say can be done?

The commission's operative claim, quoted verbatim in its findings, is that a “realistic route to act” exists for each of the 17 threats. This is the assessment's answer to the standard objection against long-horizon risk reports: that they describe problems no institution owns.

By asserting an actionable path for every listed threat, the commission shifts the burden onto governments, funders and international bodies to explain why those paths remain untaken. Aid cuts illustrate the mechanism most directly — the threat is a policy choice, reversible by the same governments making it.

For malicious AI and nuclear conflict, the actionable routes run through arms-control-style instruments: verification, treaty architecture, and restrictions on the most dangerous capabilities. The commission's inclusion of both in a health assessment functions as an argument that these belong in the same governance conversation as pandemics and famine, not in a separate security silo.

Why does this matter now?

The assessment lands amid three concurrent pressures the commission could not have ignored. Wealthy countries are contracting aid budgets. Climate negotiations continue without binding enforcement. AI capability development outpaces treaty discussion by years.

Placing malicious AI at the top of a Lancet threat list gives regulators and public-health agencies a citable, medical-franchise endorsement of the position that AI risk is population-scale risk. That citation matters in policy settings where AI safety arguments, raised by labs or activists alone, are treated as interested; a health commission's assessment carries a different institutional signature.

The 2100 deadline converts the report into a planning document. Institutions that set decadal budgets — health ministries, UN agencies, sovereign research funders — now have a named commission, a named threat count, and a named endpoint year to organize against.

The commission's closing position is the one most likely to be quoted forward: for all seventeen threats, the realistic route to act already exists. The open question the assessment leaves with its readers is whether the actors holding those routes will use them within the 75 years the commission says remain.