Most Americans want AI development to slow down or stop entirely, new poll finds

The most striking figure from this Quinnipiac University poll is that 77% of Americans want to slow or pause AI development until its safety is verified. Broken down, 47% want to slow the pace of development, 30% want a full stop, and only 5% want to speed up. These numbers show that the center of gravity in public sentiment is not 'faster' but 'confirm safety first.'

It should be noted that the source does not clearly state whether the 77% figure is simply the sum of 47% and 30%. A more cautious reading is therefore: those who favor slowing down and those who favor a full stop together form a clear majority, while those who favor acceleration are a very small minority. This structure itself is more noteworthy than any single percentage.

Echoing the demand to 'slow down' is a trust gap toward AI company leadership: 74% of respondents have little or no trust in AI company leaders. Chetan Jaiswal of Quinnipiac University interpreted this by saying that these results together indicate the public wants safeguards that do not rely solely on trusting what company leaders say.

In the same direction, 86% of respondents support independent safety standards for AI companies currently under discussion, even if this slows progress; 91% consider U.S. safeguards for AI systems important. Putting these two figures together with the 74% distrust, one can infer (this is an inference): the public tends to entrust safety responsibility to mechanisms outside companies rather than to corporate leaders' self-commitments. But the source does not explain the specific content of these 'independent safety standards,' nor does it define who verifies 'safety is verified' or by what standards.

The concerns in the poll do not remain only at an abstract level. 73% of respondents worry AI could eventually threaten human survival; 53% believe AI will bring more harm than good in daily life. The former points to long-term, extreme risk; the latter points to a real feeling that has already entered daily life.

This concern also lands in concrete space: 72% of respondents oppose building AI data centers in their own community. In other words, reservations about AI are expressed not only in 'whether to continue development' but also in 'whether related facilities should be built near me.'

On risk attribution, 52% of respondents worry more about human misuse of AI than about autonomous AI. This suggests the center of public anxiety leans more toward human misuse than toward machines going out of control on their own—though 73% still worry AI could threaten human survival. The two are not contradictory but are concerns at different levels coexisting.

There is another easily overlooked figure in the poll: 69% of respondents consider it important for the U.S. to keep up with China in AI. This creates a tension with the demand to 'slow or pause'—the public wants safety first but does not want to fall behind in AI competition.

The source does not explain how respondents would trade off between these two goals, nor whether this competition concern would weaken support for slowing development. Therefore, here one can only confirm that both attitudes exist simultaneously, and cannot infer that the public has already given a priority ranking.

What can be confirmed are the basic parameters of the survey itself: the poll surveyed 1,202 American adults from September 24 to 27, 2026, with a margin of error of plus or minus 3.5 percentage points. This means some proportions close to the margin of error have room for fluctuation in their precise values.

The boundaries are equally important. The evidence comes from only a single source, namely THE DECODER's report on the Quinnipiac University poll; it does not provide the original poll report, nor does it include the full original questionnaire; the source does not provide details such as respondent demographics or sampling method. The poll covers only American adults and cannot be generalized to other regions; it reflects public opinion in late September 2026, not a long-term trend.

In addition, the source does not state whether this poll is consistent with other poll results, nor does it mention AI company leaders' reactions to the poll. Therefore, whether these numbers can translate into policy or corporate behavior remains unknown.

Stringing the above numbers together, the most coherent interpretation is: most American adults do not treat 'faster' as the goal, but demand that AI development be premised on verifiable safety; they support independent safety standards and do not trust corporate leaders' self-assurances.

But the strength of this interpretation is limited by the evidence itself: it comes from a single poll in a secondhand report, lacks the original questionnaire and sampling details, and the verifying body and standards for 'safety is verified' are not defined in the source. Therefore, this set of numbers is better read as a clear public opinion signal rather than an already formed policy consensus.