Roman Yampolskiy, who runs the Cyber Security Lab at the University of Louisville, puts the odds of AI wiping out humanity above 99%. Meta’s Yann LeCun puts them below 0.01%. That gap, between two credentialed researchers, is where the question sits in 2026. This piece collects what surveys, prediction markets, and polls actually report about whether AI will take over the world.
Will AI Take Over the World — TL;DR
- 53% of Americans think it is somewhat or very likely that AI will destroy humanity someday (Yahoo/YouGov, October 2025).
- The median machine-learning researcher puts the chance of an extinction-level outcome at 5% (2023 Expert Survey on Progress in AI).
- Named expert estimates run from below 0.01% (Yann LeCun) to above 99% (Roman Yampolskiy).
- Metaculus forecasters moved their median date for the first general AI system from July 2031 to November 2033 over 2025.
- The Future of Life Institute’s October 2025 call to ban superintelligence drew Geoffrey Hinton, Yoshua Bengio, and five Nobel laureates.
The short answer: no current system can take over the world, and the people who study this for a living disagree sharply about whether one ever will. Researcher surveys cluster around a low single-digit chance of catastrophe, the public runs far more pessimistic, and timelines for the capability that any takeover would require moved further out last year. The numbers track this scale of AI market growth and adoption alongside rising unease.
What Are the Odds AI Will Take Over the World? Expert p(doom) Estimates
“p(doom)” is the term researchers use for the probability of an existential catastrophe from advanced AI. Individual estimates from named scientists vary so widely that no single figure speaks for the field.
| Expert | Role | p(doom) |
|---|---|---|
| Yann LeCun | Former chief AI scientist, Meta | <0.01% |
| Vitalik Buterin | Co-founder, Ethereum | ~12% |
| Geoffrey Hinton | 2024 Nobel laureate | 10–20% |
| Dario Amodei | CEO, Anthropic | 10–25% |
| Yoshua Bengio | Turing Award winner | ~20% |
| Paul Christiano | Head of research, US AI Safety Institute | ~50% |
| Roman Yampolskiy | AI safety researcher, U. of Louisville | 99%+ |
Source: P(doom) compilation, Wikipedia; TechRadar
The named figures span almost the entire range, from effectively zero to near-certainty. Hinton gives a 10–20% all-things-considered figure but says his own independent impression sits above 50%. Amodei reaffirmed a 25% chance that things go very badly in September 2025.
The spread is not noise around a hidden consensus. A 2025 arXiv survey of 111 AI experts found they split into two camps, one treating AI as a controllable tool and one treating it as an uncontrollable agent, which is why the estimates pull so far apart.
What Do Surveys of AI Researchers Show?
Single quotes grab headlines; large structured surveys give a steadier reading. These poll hundreds to thousands of practitioners rather than the loudest voices.
| Survey | Sample | Key finding |
|---|---|---|
| 2023 Expert Survey on Progress in AI | 2,778 researchers | Median 5% chance of extinction-level outcome; mean 14.4% |
| 2023 Expert Survey (catastrophe framing) | 2,700+ researchers | 58% saw at least a 5% chance of extremely bad outcomes |
| arXiv survey of AI experts (2025) | 111 experts | 78% agreed researchers should be concerned about catastrophic risk |
| arXiv survey of AI experts (2025) | 111 experts | Only 21% had heard of “instrumental convergence” |
Source: Grace et al. (2023/2024); Field, arXiv (2025)
The median researcher in the largest survey put the chance of an extinction-level outcome at 5%, while the mean reached 14.4%, pulled up by a minority of high estimates. A separate reading of the same data found nearly 58% saw at least a 5% chance of an extremely bad outcome.
Agreement that the topic deserves attention runs high at 78%, yet familiarity with specific safety ideas is low. Only 21% recognized instrumental convergence, the prediction that advanced systems pursue self-preservation as a sub-goal.
When Could AI Be Able to Take Over the World?
Any takeover presupposes a capability that does not yet exist. Timelines for artificial general intelligence are the closest measurable proxy, and 2025 was the year they stopped shrinking and started moving back.
| Source | Forecast | Date / probability |
|---|---|---|
| Metaculus (start of 2025) | First general AI system | July 2031 |
| Metaculus (early 2026) | First general AI system | November 2033 |
| Metaculus (Feb 2026) | AGI by 2029 | 25% probability |
| Metaculus (Feb 2026) | AGI by 2033 | 50% probability |
| Metaculus (2020 baseline) | First general AI system | ~50 years out |
Source: Metaculus, via 80,000 Hours (2026)
Over 2025 the Metaculus median for the first general AI system moved out by about two and a half years, from July 2031 to November 2033. As of February 2026 the same forecasters averaged a 25% chance of AGI by 2029 and 50% by 2033, with both figures up by two years over the prior twelve months.
The shift was broad. The Metaculus community, Dario Amodei, and forecaster Peter Wildeford all pushed timelines later across the year, with only one tracked forecaster moving sooner. The same period saw rapid uptake of AI coding systems such as AlphaCode, even as general capability stayed years off.
How Worried Is the Public That AI Will Take Over the World?
Public opinion runs more pessimistic than researcher surveys. Across multiple 2025 polls, majorities voice concern about long-term risk while still finding the tools useful day to day.
| Poll | Date | Finding |
|---|---|---|
| Yahoo/YouGov | Oct 2025 | 53% think AI is likely to destroy humanity someday |
| YouGov | Dec 2025 | 77% concerned AI could pose a threat to humanity |
| Pew Research Center (US) | Jun 2025 | 50% more concerned than excited about AI in daily life |
| Pew Research Center (US) | Jun 2025 | 57% rate AI’s societal risks as high |
| FLI signatories survey | Oct 2025 | 64% want superhuman AI suspended or banned until proven safe |
Source: Yahoo/YouGov; YouGov; Pew Research Center; Future of Life Institute
A majority of Americans, 53%, now think AI is likely to destroy humanity at some point. A December 2025 YouGov poll put concern that AI could threaten humanity at 77%, including 39% who were very concerned.
Pew’s share of US adults more concerned than excited reached 50% in June 2025, up from 37% when first asked in 2021. The same survey found 57% rate societal risks as high against 25% who rate the benefits as high.
The pessimism sits alongside satisfaction: 85% of Americans who have used AI chatbots found them at least somewhat helpful. Globally the mood is calmer, with a Pew median of 34% across 25 countries more concerned than excited. Tools like Midjourney’s growing user base show how fast everyday adoption has spread.
What Are Experts Doing About AI Takeover Risk?
Concern has turned into organized action rather than commentary alone. The clearest 2025 signal was a coordinated call to halt the most advanced development.
| Action | Date | Detail |
|---|---|---|
| FLI Statement on Superintelligence | Oct 2025 | Hundreds of signatories including Hinton, Bengio, Wozniak, 5 Nobel laureates |
| FLI accompanying survey | Oct 2025 | 64% of 2,000 US adults back suspension or ban |
| WEF Future of Jobs | 2025 | 92M jobs displaced, 170M created by 2030 |
| WEF Future of Jobs | 2025 | 41% of employers plan workforce reductions tied to AI |
| IMF assessment | 2024 | ~40% of jobs globally exposed to AI; ~60% in advanced economies |
Source: Future of Life Institute; World Economic Forum Future of Jobs Report 2025; IMF
The Future of Life Institute’s October 2025 statement, a single sentence calling for a prohibition on superintelligence until there is scientific consensus it can be built safely and strong public buy-in, drew an unusually broad coalition: Hinton, Bengio, five Nobel laureates, and hundreds of others. Its companion survey of 2,000 US adults found 64% want such efforts suspended until proven safe or banned outright.
Where “takeover” is measured in jobs rather than extinction, the WEF projects a net gain of 78 million jobs by 2030 despite 92 million displaced, and reports 41% of employers plan AI-driven cuts within five years. The IMF found roughly 40% of jobs worldwide are exposed to AI, rising to about 60% in advanced economies. Enterprise uptake of platforms like Azure OpenAI and GitHub Copilot is part of that shift, built on the same cloud computing buildout driving compute demand.
The Bottom Line on Will AI Take Over the World
The data does not support one answer; the disagreement is the answer. The largest researcher survey puts the median chance of an extinction-level outcome at 5%, the public puts the odds of eventual destruction above even at 53%, and named experts range from under 0.01% to over 99%.
Timelines for the capability any takeover would require moved further away in 2025, not closer, with the Metaculus median for general AI now at November 2033. What the numbers show clearly is rising concern, organized expert pushback, and measurable economic disruption already underway, against deep uncertainty about the long-term outcome.
Will AI Take Over the World — FAQs
Can AI take over the world right now?
No. No current system has the general capability for autonomous control. Forecasters on Metaculus place the median arrival of a first general AI system at November 2033, and even that does not imply takeover.
What is the expert probability that AI destroys humanity?
The 2023 Expert Survey of 2,778 researchers found a median 5% chance of an extinction-level outcome and a mean of 14.4%. Named individuals range from below 0.01% to above 99%.
How many people think AI will destroy humanity?
A Yahoo/YouGov poll in October 2025 found 53% of Americans think it is somewhat or very likely AI will destroy humanity someday. A December 2025 YouGov poll put broader threat concern at 77%.
When will AGI arrive?
As of February 2026, Metaculus forecasters averaged a 25% chance of AGI by 2029 and 50% by 2033. Both figures rose by two years over 2025, reflecting later timelines across the field.
Will AI take jobs instead of taking over?
The WEF projects 92 million jobs displaced and 170 million created by 2030, a net gain of 78 million. The IMF estimates about 40% of global jobs are exposed to AI, near 60% in advanced economies.
Citations
https://futureoflife.org/press-release/prominent-scientists-faith-leaders-policymakers-and-artists-call-for-a-prohibition-on-superintelligence/ https://80000hours.org/2025/03/when-do-experts-expect-agi-to-arrive/ https://www.yahoo.com/news/article/poll-most-americans-think-ai-will-destroy-humanity-someday-212132958.html https://en.wikipedia.org/wiki/P(doom)
