Probability of a Global AI Catastrophe
Probability of a Global AI Catastrophe
Experts put the risk of a global AI catastrophe before 2100 at 5%.
Hover to highlight a scenario, click to show or hide itTap a scenario to show or hide it
Our Analysis
How do we interpret these results
All participant groups forecast that AI catastrophic risk rises with the assumed pace of AI progress. Unconditionally, the median expert assigns a probability of 0.3% by 2030, 2% by 2050, and 5% by 2100. Under Slow Progress, the corresponding expert medians are 0.08%, 1%, and 2%; under Rapid Progress, they rise to 1%, 5%, and 10%. The public consistently assigns higher probabilities than both experts and superforecasters, with unconditional medians of 1%, 3%, and 7%, and a median of 15% by 2100 under Rapid Progress. Superforecasters are more optimistic than experts, assigning unconditional medians of 0.1%, 0.88%, and 2.4%. Differences between experts and the public are statistically significant across all reported dates and scenarios. Expert-superforecaster differences are also statistically significant in the unconditional, moderate-progress, and rapid-progress forecasts at all three dates, while slow-progress forecasts are much closer. Dispersion remains wide even among experts: under Rapid Progress, the expert interquartile range is 2%-10% by 2050 and 3%-22% by 2100.
Forecasts by scenario
Median forecasts from experts for each scenario
| scenario | 2030 | 2050 | 2100 |
|---|---|---|---|
| Unconditional | 0.3% | 2% | 5% |
| Slow Progress | 0.08% | 1% | 2% |
| Moderate Progress | 0.5% | 2% | 5% |
| Rapid Progress | 1% | 5% | 10% |
Click any cell to compareTap any cell to compare
* Denotes non-overlapping bootstrapped confidence intervals between group aggregates.
Rationale analysis
The main lines of reasoning driving the spread in estimates
Pace of progress as a risk factor
A clear majority of both optimists and pessimists view faster progress as raising risk because it compresses the time available for adaptation: “Faster progress both pulls forward the expected date of a potential catastrophe and increases its likelihood at any time, because we have less time for policy and society to respond.” Pessimists tend to focus exclusively on that factor, whereas optimists often either acknowledge the compression issue will raise overall risk but point to mitigating factors—“progress could either support more safety (a more advanced system should have better defenses against harmful uses) while also enabling more risk”—or, in a handful of cases, argue that rapid progress is a less risky path relative to the other scenarios: “It may be slow progress we most need to worry about, when models are already susceptible to adversarial attacks by people but aren't sufficiently developed to prevent them.”
Safety mandates
Many optimists argue that governments and institutions, catalyzed by smaller warning events, are likely to intervene before the 10%-mortality threshold is in danger of being reached: “Prior to such a large event, there would very likely be smaller ones, perhaps in which the fatalities are in the thousands or tens of thousands. Such events would most likely bring on a backlash in which remedies were put into place to avoid future events. (See Chernobyl, Three Mile Island, and Fukashima.)” Pessimists typically are far more skeptical: “The political willingness to effectively regulate AI at the national level or cooperate internationally is sorely lacking”; “[AI]-race dynamics within and across countries, unethical or stupid people in charge, the push to put AI in weapons…preclude both serious thinking on AI alignment and worldwide regulations on AI.”
The 10% threshold
Optimists treat ~800 million deaths as nearly disqualifying: “All of WWII, including civilian deaths, was 3%…So the question is how many nines of unlikely?”; “I just see this scenario as the stuff of Hollywood, but not realistic, especially when we consider we’d be talking about almost one billion deaths.” Pessimists often argue the high bar suggested by history is deceptive given that “our growing power to do damage is likely to outweigh” our resilience. Some also argue that the base rate, however low, is still meaningful: “The only clear qualifying event for the global population is the Black Death, but there are a few cases that seem like they’d qualify if you considered only ‘the known world to the population affected’ (the Plague of Justinian, Cocoliztli, New World smallpox). I get a rough base rate of 0.05% to 0.3% per year…It’s notable that several of these are cases of populations being exposed to a new pathogen they weren't adapted to.”
Threat-vector pathways
Optimists and pessimists largely agree that engineered pathogens, AI-misalignment, autonomous weapons, and nuclear escalation are the most likely pathways. But pessimists tend to find at least one channel plausible: “Too many possibilities for my taste. And a few years from now, ANY major catastrophe will be AI-related.” While optimists are more skeptical: “After Covid-19, the world is better equipped to deal with a pandemic. Other than a pandemic, a nuclear war could potentially wipe out that many people, but human control prevails and it would need a very serious chain of misalignment problems…(see the Swiss cheese model—a lot of independent security holes need to be present at the same time).”
Time-horizons
Most respondents deem the odds of a catastrophe occurring by 2030 to be exceptionally low, with many pointing to the fact that the slow/moderate/rapid AI-progress scenarios only speak to the state of progress by 2030. As one writes, “The 2030 resolution effectively corresponds to a single 5-year window (~2026–2030), during which AI capability is still developing.” A frequent corollary—used to justify convergence across scenarios at later dates—is that the state of AI progress by 2030 “[doesn’t] indicate much about the world or state of AI in 2100.”
Rationale examples
High-quality rationales from forecasters across the range
Optimists
More than 10% of the global population dying within five years would require a catastrophe on a scale far beyond any modern war or disaster. Even World War II, often considered the deadliest conflict in history, killed a much smaller share of the world's population. While AI could increase risks related to cyberattacks, military systems, biotechnology, or other dangerous capabilities, I do not view AI itself as a plausible primary cause of an event that kills over 10% of humanity within the forecast horizon. The most plausible pathways to losses on that scale would likely involve human decisions, state conflict, nuclear weapons, or biological weapons, even if AI played a supporting role.
For AI to be the proximate cause of an event at that scale, the vast institutional protections we now have in place would have to fail. That is, some combination of nuclear command and control, biosecurity, public health, military chains of command, and international cooperation would have to fail. These institutions may not be perfect, but they have held together since the end of WWII. We haven't seen anything remotely close to this not because the technology became safer—it didn't. Social institutions held.
Despite the fact that we are definitely playing with fire, I just don't see the risk probabilities as being…high in the near term (and I consider 2100 the near term)...Were something to go terribly wrong, we're not exactly defenseless, and 800 million people is a lot of people for AI to kill even if a system was inclined to do so. Sure, cognitive ability is a huge advantage in lethal confrontations between competitors, but it's not always determinative. And we humans are dispersed all around the world, have adapted to live in harsh environments, and fight back with vigor when threatened. We're lousy at dealing with theoretical problems but can be creative, decisive and ruthless when dealing with problems that threaten our immediate existence.
[Initially] my highest risk [was] associated with misaligned actions by a single powerful AI. Now I think a multipolar AI world is more likely. That mostly replaces the caprices of a single ASI with likely more frequent but lower impact events from multiple ASIs…My intuition is that risk drops in a multipolar world…This might effectively amount to stronger guardrails.
Pessimists
The mechanism that pushes me away from ‘this is impossible’ is not some cartoon version of an evil superintelligence suddenly deciding to purge humanity. It is that AI radically lowers the barrier between ambition and capability while doing almost nothing to improve judgment, wisdom, restraint, or systems-thinking maturity in the humans deploying it. I have personally seen people with little or no experience building secure production-grade systems wiring extraordinarily dangerous capabilities into GenAI applications because coding assistants make the implementation feel deceptively easy. I’ve seen proposals for ‘skills’ that expose unrestricted access to corporate email inboxes, plaintext workspace tokens embedded directly into tooling flows, and sensitive credentials being traced, logged, or persisted in ways that would make any experienced security engineer visibly pale. None of the people involved were malicious. That is precisely the problem. The danger is not primarily evil actors. It is confident and underqualified actors suddenly possessing the ability to build systems far beyond their actual level of engineering maturity. GenAI dramatically compresses the distance between ‘I have an idea’ and ‘this is now running in production connected to real infrastructure.’ Historically, a lack of technical capability acted as a natural governor on many bad ideas. AI-assisted software generation weakens that governor substantially.
The most plausible pathway to AI-driven catastrophe is not a rogue superintelligence acting alone, but the familiar logic of competitive dynamics pushing rational actors toward collectively irrational outcomes. Nations racing for AI supremacy face a classic prisoner’s dilemma: each has strong incentives to automate weapons systems and decision-making, even knowing that universal automation increases everyone’s risk. First-mover advantages in autonomous weapons, as illustrated vividly by the proliferation of AI-guided drone swarms, mean that actors who fall behind may rationally contemplate preemptive strikes before a capability gap becomes insurmountable, implying that countries surging ahead also have the incentive for a preemptive strike. Compounding this is the speed problem: even well-intentioned actors may lose the ability to de-escalate once AI-enabled conflict begins, since crises could unfold in hours rather than the days that diplomatic resolution historically requires.This dynamic is further compounded by the democratization of destructive capability: open-weight models dramatically lower the barrier for small groups or individuals to engineer biological or chemical weapons, meaning the number of actors capable of mass casualties grows faster than the institutions designed to contain them.
Attribution to AI for e.g. suicide epidemics, or increases in depression, or unemployment/poverty/inequality, civil wars (all statistically…AND causally linked to AI already—via AI psychosis, toxic AI companionship, fraud enablement, automation/shift to capital-intense means of production, polarisation...) could get you to 10% pop over 5 years, even without a single traceable cause.