ABC/Post/Ipsos Poll: 57% of Americans Say AI Is Bad for Society, 66% Oppose New Data Centers

A new national poll adds hard numbers to a mood that data-center fights and midterm campaigns have already shown. In the ABC News/Washington Post/Ipsos poll published October 21, 2026, 57 percent of Americans say AI is bad for society, 66 percent oppose building new data centers in the US, and 72 percent oppose one in their own community. The poll was fielded October 9 to 13 among 2,545 US adults on Ipsos's probability-based KnowledgePanel, with a margin of error of plus or minus 2 points.

For people who build with AI, the headline is less important than the shape underneath it. Americans are split evenly on whether AI helps them personally, yet lopsided on whether it is good for society and on whether its physical footprint should be built next door. That gap is where policy, local politics and product trust are now being decided.

The poll at a glance

Who thinks AI is bad for society

Age matters, and the direction may surprise people who assume young users are the enthusiasts. Per the poll, 65 percent of under-30s, 59 percent of those aged 30 to 39 and 53 percent of those 40 and older say AI is bad for society.

Party splits are sharp but not absolute. Sixty-three percent of Democrats and 58 percent of independents say bad, while Republicans lean the other way, 51 percent good to 47 percent bad. The personal-impact question reverses the pattern: 57 percent of Republicans say AI is good for them, while 55 percent of Democrats say it is personally bad. Independents split 50 to 49.

The one-line read: personal benefit is contested, collective worry is majority. That is the profile of a technology people use and distrust at once.

What people fear: hacking leads the list

The poll asked about specific risks, and the order is informative.

Hacking tops the list, ahead of jobs. That tracks the news cycle: we have covered Anthropic's study of a self-replicating AI botnet and the push for tools that find bugs before attackers do. Extinction, the topic that dominates some online debate, is the least-cited concern at a third of respondents. Decision-making dependence is notable: 62 percent overall, and 67 percent of under-50s, worry about losing the ability to decide without AI.

Benefits poll far lower. Just 34 percent see a big chance of AI making everyday tasks easier, 30 percent a big chance of US dominance in AI technology, and 20 percent a big chance of making people more productive. Only 13 percent see a big chance of AI creating many new jobs, while 52 percent say there is little or no chance.

Data centers: opposition is wider than the AI debate

The data-center numbers are the most actionable. Nationally, 66 percent oppose new data centers and only 30 percent support them. Support rises to 53 percent among MAGA Republicans. When the question becomes local, opposition rises to 72 percent.

Opposition crosses party, age and geography. Rural residents, who are most likely to be offered a site, are the most opposed.

Rounded houses with a blank protest sign, illustrating local opposition to a data center in the community

How the poll lines up with events on the ground

The results arrive amid a run of local fights that we have been tracking.

• San Francisco's 45-day data center moratorium and the New York debate in the Hochul and Ocasio-Cortez data center moratorium explainer show elected officials moving toward pauses.
• In India, Amazon's Thane data center drew protests over water and power.
• Companies are adjusting tactics: Amazon dropped data-center NDAs, and Google stopped its NDA practice with local officials, per reports we covered.
• On cost, a Senate investigation alleged hyperscalers misled the public on data-center costs.
• In Kansas, Digital Realty's De Soto project and a $1,000-per-household claim show the offers communities are weighing.

A 72 percent local-opposition number suggests these tactics, which are mostly about transparency and community payments, are working against a very strong headwind. The poll does not test which sweeteners change minds, so it cannot say whether dividends or disclosure move the needle.

Regulation and politics

Two-thirds say the federal government is not doing enough to regulate AI. That includes 80 percent of Democrats, 66 percent of independents and 54 percent of Republicans. A majority of Republicans wanting more regulation matters in a party whose leadership has favored a lighter touch.

Two-thirds also disapprove of President Trump's handling of AI, while 30 percent approve. Among MAGA Republicans, about six in ten Republicans, 80 percent approve; non-MAGA Republicans split 49 to 50. Democrats disapprove at 92 percent and independents at 75 percent.

Neither party owns the issue. Forty-nine percent trust neither, 26 percent trust Democrats more and 19 percent Republicans. In April, according to ABC, Americans were roughly split between the parties, with about half again trusting neither. Our coverage of Democratic AI regulation messaging ahead of the midterms and of Trump's claim that data centers must supply their own electricity shows both sides positioning on exactly the issues the poll measures.

Reading the poll carefully

Polls are snapshots, and a few cautions apply.

• Question wording drives results. "Bad for society" is a broad judgment, not a statement about any product people use. Many respondents likely use AI tools daily.
• Half-sample questions have larger error. Some items went to about 1,272 adults with a 2.8-point margin, and subgroup figures such as "rural" or "under 30" carry wider margins than the headline.
• Support and opposition are not intensity. Seventeen percent say AI is extremely bad for society; only 2 percent say extremely good. That asymmetry in strong feeling may matter more than the 57-42 split.
• Opposition to a data center is not opposition to AI. Noise, water, power bills and land use drive local fights regardless of attitudes toward the models inside.
• Correlation with action is uncertain. A poll does not show who will vote, organize or attend a zoning hearing.

What this means for people building with AI

• Expect trust friction. If most of your users think AI is bad for society, product claims need evidence, and safety, security and transparency features are selling points rather than overhead.
• Plan for infrastructure politics. If you depend on new capacity, sites and power deals will face organized local resistance. Timelines slip when communities say no.
• Security messaging matters. With hacking the top-ranked risk at 73 percent, teams shipping agents should show how they limit dangerous actions.
• Watch regulation. A two-thirds majority for more regulation, including a majority of Republicans, raises the odds that state and federal proposals keep advancing.
• Job claims draw scrutiny. With 65 percent expecting many jobs eliminated and only 13 percent expecting many created, claims about AI and employment get fact-checked by an audience already skeptical.

Why the personal-versus-society gap matters

The most revealing pair of numbers is 50 percent personally positive against 42 percent socially positive. People can experience a tool as useful in their own workflow while doubting what it does in aggregate: to jobs, to attention, to electricity bills and to security. Product teams often read adoption figures as approval, but this poll shows the two can diverge. A user who relies on a chatbot every day can still tell a pollster, and later a county commission, that the technology is bad for the community.

That divergence also explains why local politics may move faster than national politics. A federal bill needs a coalition across a divided Congress, but a zoning board needs a handful of votes and a room of neighbors. With 72 percent opposed to a data center nearby, and rural residents at 78 percent, the people closest to a proposed site are the least persuadable, which raises the cost and the timeline of new capacity.

What the poll does not tell us

It does not say how views differ among regular AI users and non-users, how opinion has changed over a longer span beyond the April party-trust comparison, or what would shift local opposition. ABC's write-up gives the headline results and breakdowns, but readers wanting the full toplines should look for the Ipsos and Washington Post releases. We have not seen the full questionnaire, so exact wording for each item should be checked there.

Related reading

• San Francisco 45-day data center moratorium
• Hochul and AOC data center moratorium explainer
• Amazon Thane India data center protests
• Senate investigation: hyperscalers and data center costs
• Democrats, AI regulation and the midterms
• Trump on data centers supplying their own electricity
• Amazon drops data center NDAs
• Official: ABC News poll report and Ipsos

Figures are from ABC News's report of the poll, published October 21, 2026.