Why public opinion belongs in a manual about AGI
Everything else in this manual asks what AI systems can do and what labs and researchers say about where that is headed. This page asks a different question: what does the public — the people who will actually live through whatever AGI turns out to mean — believe about the technology right now, and how has that belief moved? Public opinion is not evidence about AGI timelines or capability, the way a benchmark score or a deployment incident is. It is evidence about something else that matters just as much for how this transition actually goes: the level of public consent, trust, and tolerance that companies, regulators, and researchers are operating with. See how to weigh claims about AGI for why a survey result and a capability claim need to be read differently — a poll tells you what people believe, not whether the belief is correct.
The pattern across four independently fielded surveys — Pew Research Center, Gallup, Edelman, and Ipsos/Stanford HAI — is consistent enough to take seriously: concern is rising in the countries most exposed to frontier AI development, trust in the institutions building and deploying it is falling, and a wide gap separates how China’s public feels about AI from how the US, UK, and much of Western Europe feels.
The US public has grown more concerned, not less, since 2021
Pew Research Center’s American Trends Panel survey, fielded June 9–15, 2025 among 5,023 U.S. adults and published September 17, 2025, found that 50% of Americans are now more concerned than excited about AI’s growing role in daily life, versus just 10% more excited than concerned (38% feel both equally). That concern share has moved substantially since Pew first asked the question in 2021, when only 37% said they were more concerned than excited. Four years of rapidly improving models, in other words, have not made the American public more comfortable with AI — the opposite has happened.
The same survey found 57% of Americans rate AI’s risks to society as high, against only 25% who rate its benefits as high — a wide and asymmetric gap between perceived downside and perceived upside. Younger adults are, if anything, more worried than older ones: among 18–29-year-olds, 55% say they’re more concerned than excited, compared to 11% who are more excited than concerned. Application matters too — the same respondents are broadly comfortable with AI in specific, bounded roles (74% support its use in weather forecasting, 66% accept it in medicine development) while opposing it in others (73% oppose AI weighing in on religious questions), which suggests the concern is not a blanket rejection of the technology so much as a judgment about where it does and doesn’t belong.
A 25-country survey shows the US isn’t an outlier — but isn’t typical either
Pew’s companion global survey — 28,333 adults across 25 countries outside the US, fielded January 8–April 26, 2025, plus separate US samples, published October 15, 2025 — puts the median country at 34% more concerned than excited, 42% equally concerned and excited, and only 16% more excited than concerned. The US sits among the more anxious countries in this set, alongside Italy, Australia, Brazil, and Greece, each with roughly half their population “mainly concerned.” South Korea is the clear outlier at the other end, with only 16% mainly concerned — a reminder that “global opinion on AI” is not one number but a spread shaped by each country’s own relationship to the technology, its economy, and its government. See AGI timeline forecasts for the parallel spread among researchers and labs themselves: public concern and expert timeline estimates are both wide distributions, not single agreed-upon figures, and it would be a mistake to treat either as more settled than it is.
Trust in who regulates AI follows a similarly uneven pattern. Across the countries Pew surveyed, the median respondent trusts the EU’s regulatory approach to AI more (53%) than either the US’s (37%) or China’s (27%) — though trust in one’s own country’s regulatory capacity varies enormously, from 89% in India and 74% in Indonesia down to 22% in Greece. That gap between how much people trust a government to build AI and how much they trust it to govern AI responsibly connects directly to the concerns raised in civil rights, discrimination, and surveillance — regulatory trust is, in large part, a bet on whether oversight will catch misuse before it does harm at scale.
Trust in AI is declining further, not stabilizing
Gallup’s most recent tracking survey, “Americans Cool Toward AI,” fielded May 4–11, 2026 among 3,270 U.S. adults (published July 28, 2026), shows the erosion continuing into 2026. Only 27% of Americans trust businesses “a lot” or “some” to use AI responsibly, down from 31% a year earlier — and the decline is sharpest among 18–29-year-olds, whose trust fell from 30% to 20% year-over-year. The share who believe AI does more harm than good rose to 39%, up from 31% in 2025 (among young adults specifically, that figure jumped from 36% to 47%). Separately, 79% of Americans now anticipate AI will reduce the number of US jobs over the next decade, up from 73% the year before, with concern among 18–29-year-olds rising 13 points to 75%. Two consecutive years of Gallup data moving in the same direction is a genuine trend, not survey noise — trust in the institutions deploying AI is going down even as the technology itself becomes more capable and more widely used.
The China–West trust gap is large and consistent across pollsters
The most striking finding across this research is not just that trust in AI differs by country, but by how much. Edelman’s five-country Flash Poll — Brazil, China, Germany, UK, and US, released November 19, 2025 — found trust in AI at 87% in China against just 32% in the United States (36% in the UK, 39% in Germany, 67% in Brazil). The generational pattern holds inside each country too: among 18–34-year-olds, 88% of Chinese respondents trust AI versus 40% of young Americans. Only 17% of Americans say they “embrace greater use of AI,” compared to 54% of Chinese respondents.
Ipsos’s cross-country data, drawn from a 2024 fielding of 23,685 adults across 32 countries and reported in Stanford HAI’s 2025 AI Index, shows the same East–West divide from a different angle: the share who believe AI products and services offer more benefit than drawback runs 83% in China, 80% in Indonesia, and 77% in Thailand, against just 40% in Canada, 39% in the United States, and 36% in the Netherlands. That gap is real but not frozen — since 2022, optimism has risen in several Western countries too (Germany and France each gained 10 points, Canada and Great Britain 8 points, the US 4 points), suggesting the low end of the gap is narrowing slightly even as the high end holds steady.
Reading the gap honestly
It is tempting to read a China–US trust gap this wide as a simple verdict on which public has the more accurate risk assessment. That would be a mistake, and it is exactly the kind of inferential overreach this manual tries to flag rather than repeat. A trust survey measures belief, shaped by each country’s media environment, its government’s own framing of AI as a strategic asset, and its citizens’ baseline trust in institutions generally — not an independent audit of AI’s actual safety or benefit. Two things can be true simultaneously: Chinese respondents may have less exposure to Western-style AI incident reporting and civil-liberties debates that dampen US and European trust, and Western respondents may be reacting to real, documented harms that haven’t been as visible or as widely reported elsewhere. Neither survey resolves which factor dominates.
What these four data sets do establish, reliably and consistently, is that public sentiment toward AI in its most AI-exposed democracies is trending down, not up, even as capability trends up — and that whatever consensus exists about AI’s trajectory among researchers, it is not shared by the publics who will be governed by whatever comes next.