Nearly two-thirds of Americans believe artificial intelligence could one day destroy humanity, and more of them want frontier development paused than want it to continue — according to a POLITICO Poll published on 16 September 2026.
The survey found 63 percent of Americans say AI could destroy humanity one day, framed as at least a moderate risk. On the question of what to do about it, 48 percent favour pausing the development of more advanced AI models, compared with 31 percent who favour continued development. POLITICO’s own read of the numbers, quoted by Political Wire: the findings “signal an emerging tech-skeptical realignment of American politics with consequential midterm and presidential elections looming.”
That is a striking public position for a country whose consumer behaviour tells the opposite story — and the tension between the two is the real story.
The usage-trust gap keeps widening
A Pew Research Center survey released in June 2026 found 49 percent of American adults already use AI chatbots, up from roughly a third in 2024 — while 40 percent said AI would ultimately be worse for society and only 16 percent expected a positive impact over the next 20 years. In that survey, two-thirds of Americans said they had little or no confidence in the US government’s ability to regulate AI effectively, and 59 percent said the same about the companies building it. The same paradox shows up in New Zealand workplace data: our analysis of Employment Hero’s NZ survey found workers adopting AI faster than any official framework, with nearly four in ten feeling that using AI is a form of cheating — adoption and unease rising on the same curve.
The new POLITICO numbers harden that picture from “distrust” into something closer to active fear, and add a policy preference: the public is not asking for faster safety research alongside continued scaling — a plurality wants the scaling itself to stop.
The politics are shifting under the poll
What makes the poll consequential is where American politics was already moving this week. As we covered on 15 September, Congress is preparing a stack of AI guardrail bills from both parties — a Republican kill-switch proposal, Senator Sanders’ superintelligence ban — even as President Trump dismisses the risk debate as a “HOAX” and “conspiracy.” Senator Chris Van Hollen, responding to the new poll, argued the risk landscape reflects what he called a corrupt bargain between the administration and AI investors, and called for mandatory safeguards and a comprehensive testing regime before development continues.
A 63-percent-majority risk perception with a plurality pause preference is the kind of number that converts safety bills from fringe proposals into electorally safe votes — and it does so across party lines, since anxiety about AI cuts through the partisan map. POLITICO’s analysis flags exactly that: the issue is becoming a realignment question heading into the 2026 midterms.
The industry is asking for a pause too — with different motives
The public’s pause preference has an uncomfortable echo in the industry’s own leadership. In recent weeks Sam Altman delayed OpenAI’s IPO timeline over safety concerns, and Dario Amodei’s warnings about frontier-model risks drew a direct slowdown call that rattled markets. When the lab chiefs say “slow down” and two-thirds of the public says “could destroy humanity,” the overlap is real but the motives diverge: executives seek regulatory certainty and reputational cover; voters are expressing dread. The 31 percent who want development to continue regardless are, notably, outnumbered by the 48 percent who want a hard pause — yet the industry’s actual behaviour, judged by compute build-outs and release calendars, aligns with the minority position.
Whether that gap closes through legislation, through a market repricing of AI bets — as the recent AI stock slump suggested — or not at all is the open question the poll does not answer.
What the poll does not tell you
The topline is stark, but three caveats matter. First, the survey measures perceived risk (“could destroy humanity”), not expected risk — the phrasing captures belief in possibility, not prediction. Second, POLITICO has not yet published full crosstabs publicly beyond the toplines and its Trump-voter analysis, so the partisan split inside that 63 percent is not yet independently verifiable. Third, opinion polls about abstract technological risk have historically moved less than headlines imply — though the pause preference is the unusual part here, since it converts a feeling into a concrete policy demand.
FAQ
What did the POLITICO Poll find? Published on 16 September 2026, the poll found 63 percent of Americans say AI could destroy humanity one day, and 48 percent favour pausing development of more advanced AI models versus 31 percent favouring continued development.
Is this the first poll showing AI anxiety? No. Pew Research Center’s June 2026 survey found 49 percent of US adults use AI chatbots while only 16 percent expect AI to benefit society over 20 years, and two-thirds distrust government regulation of AI. The new poll’s distinctive finding is the plurality preference for actually pausing frontier development.
What is happening in Congress on AI safety? Both parties are drafting AI guardrail bills, including a Republican kill-switch proposal and Senator Bernie Sanders’ superintelligence ban, while President Trump has dismissed the risk debate as a “HOAX.” Our 15 September coverage details the bill stack.
Has the AI industry called for a pause before? Individual leaders have: Sam Altman tied OpenAI’s IPO timing to safety questions, and Dario Amodei has issued slowdown warnings. A collective industry pause — the kind 48 percent of Americans now say they favour — has never happened, and the industry’s compute investments point the other way.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
