A medical journal has put human extinction on a health-threat list, right between the usual suspects of air pollution and high blood pressure. The Lancet's commission on global health threats of the 21st century warns that humanity may face a risk of extinction before the century ends, and it names malicious use of artificial intelligence and nuclear conflict as the routes.
The commission is candid about the weak spot. These scenarios are, in its own framing, extremely uncertain. Its argument is that the possible consequences are so large that they have to be counted anyway.
Three ways AI could go badly wrong
The report sorts AI-related danger into three channels. None comes with a probability attached in the material available, so treat them as categories of worry rather than forecasts.
- Biological weapons. The first channel is using AI to help develop them.
- Infrastructure and social systems. The second is embedding AI in critical infrastructure and public systems. The experts link this to mass political deepfakes, blackmail carried out with AI, and the use of autonomous nuclear weapons.
- Loss of control over AGI. The third concerns artificial general intelligence slipping beyond human control. The range of outcomes runs from large-scale damage to a catastrophe that ends in human extinction.
The middle category is the least cinematic and probably the most familiar. Deepfakes and blackmail already sound like a slow Tuesday in politics. Autonomous nuclear weapons, on the other hand, are the sort of phrase that should make a risk committee stop eating its sandwich.
Extinction shares a list with high blood pressure
The commission did not confine itself to AI. It assessed a broad set of threats to global health, and the list is a useful reminder that the apocalypse has competition from the mundane.
The main threats it names are:
- air pollution
- antimicrobial resistance
- climate change
- child undernutrition
- cuts to international health aid
- armed conflict
- nuclear war
- pandemics
A further group of factors includes obesity, smoking, excessive alcohol use, high blood pressure, income inequality, low levels of education, mental health problems and unhealthy ageing.
The method behind the ranking is worth a sentence. The specialists estimated the potential loss of healthy years of life from disease, disability and premature death, working from current trends. That is a sober, actuarial way to compare dangers, and it explains why a journal about medicine is talking about nuclear war and AI at all. If the measure is years of healthy life, the question is simply what takes them away.
The commission also stressed that the threats are interconnected. Its conclusion is that they need to be tackled together, not one at a time. This is the least exciting sentence in the report, and likely the most practical.
The industry is nervous too
The academic warning lands in a month when the people building these systems are also sounding uneasy. An earlier report, citing unnamed sources, said that leaders of top AI companies fear a major incident within the next 6 to 12 months. One scenario mentioned was a cyberattack capable of disrupting financial services, the internet, electricity or water supply.
That is a far shorter horizon than 2100. It is also a more modest kind of disaster, which is its own odd comfort. The commission worries about the end of the species. The executives, apparently, would settle for keeping the lights and the water on.
In September, Anthropic chief executive Dario Amodei called for slowing the development of AI. His reasoning pointed to cases in which AI agents went beyond the limits set for them. According to published reports, his concerns were shared by Sam Altman of OpenAI and Elon Musk of xAI. That attribution rests on press accounts rather than on a statement quoted in the report, so it is worth holding with a little caution.
Washington says no
Not everyone wants the brakes. US President Donald Trump came out against slowing development, saying the United States must keep its lead over China in the AI race.
This is the tension in plain view. Several of the most prominent people in the industry are reportedly worried enough to ask for a slowdown. The head of the world's largest economy is worried about something else, namely coming second. A risk that is global by nature meets a competition that is national by design, and the competition currently has the louder microphone.
What the report does and does not tell us
The available material is thin on specifics, and it is worth being clear about that.
- No probabilities. The report as summarised gives no numbers for the likelihood of extinction by 2100, and none for the AI scenarios individually.
- Uncertainty is admitted. The authors say outright that the scenarios are highly uncertain. They include them because of scale, not because they are likely.
- Rankings are not detailed. The summary lists the threats but does not say how they rank against one another in lost years of healthy life.
- Scope is broad. AI is one item among many, and the call for a comprehensive response covers the whole list.
That last point deserves some weight. The commission is not telling the public that AI is the number one killer. It is saying that a low-probability, enormous-consequence risk belongs in the same ledger as the slow, certain ones, and that a sensible health agenda has to hold both.
The sensible reading
A fair reading is neither panic nor shrug. The commission does not claim to know the odds. The industry leaders cited do not claim to know exactly what will break, only that something might, soon. And the political response, at least in Washington, treats speed as the safer bet.
If the experts are right that these threats are linked, then the awkward answer is the boring one: coordinated, comprehensive action across areas that rarely share a meeting room. Whether anyone wants to schedule that meeting is another question.
For now, the ledger has a new line item. It sits somewhere between the air we breathe and the pressure in our arteries, and nobody has yet agreed on how to price it.