What does the Lancet’s 10.4-billion-DALY AI warning mean?
A new Lancet Commission estimates that action against malicious AI use could avert 10.4 billion years of healthy life lost by 2100. It is a low-probability, high-impact scenario built partly from expert judgement—not a forecast that extinction will occur.
The short version
- What happened
- The Commission estimates that measures reducing malicious AI risk could avert 10.4 billion disability-adjusted life years through 2100, but the number describes a scenario gap rather than observed harm.
- Who it affects
- Public-health teams may need to plan for AI-enabled biological misuse alongside conventional outbreak preparedness, without treating the scenario estimate as a forecast.
- What remains unproven
- The 10.4-billion-DALY figure is a scenario estimate through 2100, not observed health loss or a prediction of one event.
Keep in mind
The 10.4-billion-DALY figure is a scenario estimate through 2100, not observed health loss or a prediction of one event.
Malicious AI lacks the historical data used for conventional disease-burden modelling, so the assessment relies partly on structured expert judgement and superforecasters.
Long-horizon results are sensitive to assumptions about capability, access, safeguards, population and geopolitical conditions.
A common DALY unit aids comparison but cannot erase ethical and distributional differences among threats.
The report does not prove that current AI systems can independently execute an extinction-scale attack.
Source analysis · 3 linked sources. Primary analysed source date: 11 October 2026.
Read the full report · 8 min The evidence, methods, implications and limitations
The answer: it is a warning about an extreme scenario, not a prediction
The Lancet Commission estimates that action to reduce the malicious use of artificial intelligence could avert 10.4 billion disability-adjusted life years, or DALYs, between 2025 and 2100. A DALY represents one year of healthy life lost through illness, disability or premature death. The large number is intended to make a rare but potentially enormous hazard comparable with other threats in a common health metric.
It does not mean that researchers observed 10.4 billion years of harm, that such harm is expected, or that there is a measured 10.4-billion-DALY event on the way. The Commission describes malicious AI as low likelihood and potentially extinction-scale. Sparse historical data forced the team to use structured expert judgement and superforecaster assessments rather than the epidemiological evidence available for threats such as obesity or air pollution. The result is therefore a scenario-dependent estimate with much wider uncertainty than a measured disease burden.[1][2][3]
What the Commission assessed
The 35 commissioners span health, economics, security, governance and technology. Their report considers threats over a 75-year horizon, from 2025 to 2100, and draws on risk analysis covering 204 countries and territories. From hundreds of candidates, they selected 17 interlinked threats that could produce catastrophic health loss and for which decision-makers have a plausible route to act. The list includes established burdens, such as obesity, climate change, low educational attainment and antimicrobial resistance, alongside rare hazards such as nuclear conflict, future pandemics and malicious AI use.
A threat entered the catastrophic set when it could account for at least one billion cumulative DALYs by 2100, sufficient information existed to assess it, and policy or investment could plausibly change the outcome. The researchers compared a plausible better future with a plausible worse one. The distance between those futures is reported as amenable DALYs: healthy years that could potentially be preserved by prevention, preparedness or other action. That is different from projecting one most-likely future.[1][2][3]
Why the AI estimate is methodologically different
Many global-health models start with observed deaths, diagnoses, exposures and trends. No historical dataset records a century of catastrophic AI-enabled attacks. The Commission therefore could not estimate malicious-AI burden by extending a stable mortality series. Its methods appendix says that threats with limited historical data—including nuclear conflict, future pandemics and malicious AI—were assessed through alternative approaches, including structured expert judgement and superforecaster assessments. Those judgements combine possible impact with a view of probability, then translate the result into a health-loss framework.
This approach is useful for decisions that cannot wait for repeated disasters, but it is highly sensitive to assumptions. Small changes in the probability assigned to an extinction-scale outcome can move an expected burden by billions of DALYs. The long horizon magnifies uncertainty about AI capability, access, safeguards, population and geopolitics. The Commission calls its estimates indicative and presents a range of plausible futures. Readers should not compare the 10.4 billion figure with a clinical trial estimate as if they had the same empirical foundation.[1][2]
What 'malicious AI' means in this report
The concern is intentional use of increasingly capable systems to cause mass harm, including help in creating or deploying biological weapons. That is narrower than everyday algorithmic error and different from a claim that an AI system will independently choose to eliminate humanity. The pathway requires several links: capable models, access to dangerous knowledge or tools, a motivated actor, operational execution, failure of safeguards and a response too slow to contain the damage.
Each link is uncertain, and the Commission does not demonstrate that present systems can complete the chain. Nor does it offer a single probability of extinction that readers can treat as settled. It asks health planners to recognise that a low-likelihood pathway may still deserve preparation when the consequence could be global. The distinction matters: responsible reporting should neither dismiss the risk because evidence is incomplete nor convert a precautionary scenario into a forecast of inevitable catastrophe.[1][3]
What this means for public-health and technology leaders
For public-health services, the practical message is to connect AI governance with biosafety, surveillance and emergency readiness. Preparedness could include rapid detection of unusual outbreaks, laboratories able to characterise novel agents, resilient supply chains, tested communication plans and international reporting arrangements. Those capabilities have value against naturally occurring outbreaks and conventional biological attacks as well as any future AI-enabled event.
For model developers and providers, the scenario supports proportionate controls around biological-design capabilities: evaluations before release, tiered access, monitoring for suspicious use, secure handling of sensitive tools, incident reporting and research on whether safeguards continue to work when users combine systems. None of these measures can be declared effective from the Commission's DALY figure alone. They require their own testing, transparent failure reporting and assessment of costs, workarounds and unequal access to beneficial research tools.[1][3]
The wider global-health comparison needs care
The Commission is not a league table telling governments to divert resources from current killers to speculative technology risks. Its accompanying explanation says the threats interact and that findings should change as evidence improves. Climate, conflict, weakened health systems, inequality and declining aid can all increase vulnerability to an outbreak, whatever its origin. Conversely, investment in surveillance, vaccination and trusted institutions can reduce several risks at once.
Distribution also matters. The modelling spans 204 countries and territories, but capacity to prepare is uneven. Wealthier states and technology companies may control the most capable systems and laboratories, while people in lower-resource settings can bear disproportionate harm from disrupted health services, delayed access to countermeasures or economic spillovers. A global response must therefore include financing, access and governance beyond the countries where frontier models are built.[1][3]
Limits and what would change the assessment
The 10.4-billion-DALY estimate extends across 75 years and depends on uncertain probabilities and consequences. Structured experts and superforecasters can make assumptions explicit, but they cannot create historical evidence that does not exist. Expert groups may share blind spots, and forecasts can become outdated as models, safeguards and geopolitics change. DALYs offer a common unit but can hide ethical differences between chronic illness, local inequity and extinction. The Commission was funded by the John Stanton and Teresa Gillespie Family Foundation and Pax Sapiens, a disclosure that readers should retain alongside its institutional affiliations.
Confidence would change with publication of the AI-specific elicitation questions, participant selection, probability distributions, aggregation choices and sensitivity tests in forms that independent teams can reproduce. Empirical evaluations should track whether models materially lower barriers to designing pathogens, whether safeguards resist expert attempts to bypass them and whether monitoring catches realistic misuse without blocking legitimate work. Repeated independent forecasts, incident evidence and annual revisions could then show whether the estimated risk is rising or falling. Until then, the Commission supports preparedness under deep uncertainty—not a claim that an AI catastrophe is certain or imminent.[1][2][3]