A Lancet commission has identified 17 interconnected threats that it says could produce catastrophic losses of health worldwide by 2100, placing familiar hazards such as air pollution, malnutrition and smoking alongside nuclear conflict and the malicious use of artificial intelligence.
The assessment, described in reports published from Oct. 10 through Oct. 12, is significant less as a forecast than as an attempt to put slow-moving public-health burdens and rare, potentially civilization-shaping disasters into a common framework. It concludes that malicious AI and nuclear conflict are the only two risks in its set that could, in some scenarios, lead to human extinction. That is a statement about possible consequences, not a prediction that either event will occur.
The 17 threats span environmental, medical, social and geopolitical conditions. Reporting by EL PAÍS, CNN and a Sky News report agrees on the core finding: the commission selected 17 risks from hundreds considered because each could generate vast health losses but could also be addressed through action.
One measure for very different dangers
The commission used disability-adjusted life years, or DALYs, to compare threats. A DALY combines years of life lost to premature death with years lived in poor health or disability. Sky News reported that the commission defined a catastrophic threat as one capable of accounting for at least 1 billion DALYs.
That threshold puts the scale in perspective. One billion DALYs does not mean one billion deaths, nor does it describe a single year. It represents a cumulative loss of healthy life that can arise through mortality, illness and disability over time. The metric is widely used in global-health research, but applying it to future scenarios requires assumptions about populations, exposure, prevention and the severity of events that may never happen.
CNN reported that the analysis drew on data from the Institute for Health Metrics and Evaluation’s Global Burden of Disease work and modeled future healthy-life losses from illness, disability and premature death. The underlying approach can draw more heavily on observed disease burdens for such hazards as high blood pressure, tobacco use, obesity, child malnutrition and antimicrobial resistance. It is inherently less certain for unprecedented events, including a global malicious-AI incident or a major nuclear war.
The reported list also includes naturally occurring pandemics, climate change, conventional conflict, declining development aid and poor mental health. These threats can overlap rather than operate separately: conflict can disrupt health care and food systems, climate change can alter disease and nutrition risks, and weakened public institutions can make outbreaks harder to contain. The commission’s premise is that the interaction among hazards can amplify health damage.
AI estimates reflect assumptions, not observed harm
The AI section has drawn particular attention because it places a rapidly developing technology within a global-health risk assessment. According to EL PAÍS, the commission considered pathways including AI-assisted development of pathogens, cascading failures involving critical infrastructure, and loss of human control over misaligned autonomous systems.
Those are scenarios, not documented examples of a global-scale health event caused by malicious AI. CNN noted that no malicious-AI event of truly global magnitude has occurred and characterized the commission’s estimates of associated health losses as theoretical. Its reporting said the modeled loss could total about 10.4 billion healthy-life years worldwide by 2100, a figure that should be read as an estimate conditional on the model’s assumptions, rather than a measured burden.
Reported judgments about the likelihood of an AI catastrophe also varied sharply depending on who was asked. EL PAÍS said superforecasters consulted by the commission estimated a 2.1% probability of an AI catastrophe by 2100 and a 0.4% chance of extinction, while AI experts consulted estimated 12% and 3%, respectively. Neither set of figures is an observed probability. They are expert assessments of uncertain future events, and the gap illustrates how much the estimates depend on definitions, expectations about technological development and beliefs about safeguards.
Nuclear conflict presents a different kind of uncertainty. Its destructive effects have historical and physical foundations, but the size, location and duration of any future conflict, as well as its climatic and food-system consequences, cannot be known in advance. The commission’s inclusion of nuclear war therefore does not establish a date or probability for such a conflict; it identifies the scale of the health loss that could follow under severe scenarios.
Established burdens remain central
The report’s more speculative risks should not obscure the hazards already producing large, measurable health burdens. Air pollution, tobacco, malnutrition, high blood pressure and obesity are linked to well-documented illness and premature death. Climate change and antimicrobial resistance are also active public-health problems whose future toll will depend partly on policy, health-system capacity and unequal access to prevention and treatment.
EL PAÍS reported that the commission emphasized the differing levels of uncertainty across its list, contrasting the relatively established trajectories of smoking and malnutrition with the much less knowable prospects of AI misuse and nuclear conflict. That distinction is important for interpreting the commission’s work: the common DALY yardstick is useful for comparing potential scale, but it does not make the evidence base equally strong for every threat.
News accounts differ on the number of people on the commission, describing either 35 leaders or 37 experts. The available reports do not resolve that discrepancy. They do agree that the assessment drew on a multidisciplinary effort and that the underlying global analysis covered 204 countries and territories.
The commission’s conclusion, as reported by Sky News, was that every identified threat has a realistic route for action. The reporting does not establish a single prescribed agenda for AI or nuclear risk. But the broader public-health implication is more concrete: reducing catastrophic losses would require sustained work on the present-day burdens already tracked in health data, alongside preparations for lower-probability events whose consequences could be far larger than their apparent likelihood.