America’s enthusiasm for artificial intelligence is colliding with a much sharper demand for guardrails.
A new Reuters/Ipsos poll finds that 73% of U.S. adults believe AI companies have not gone far enough to prevent their technology from causing serious harm to society. The result is not a technical forecast that catastrophe will occur. It is a measure of public confidence—and right now, that confidence is thin.
What the poll found
The Reuters/Ipsos survey, published September 22, found concern extending well beyond one headline number:
- 73% said AI companies have not done enough to prevent serious societal harm.
- 39% said AI is having a negative effect on society, up from 36% a month earlier and the highest share since Reuters/Ipsos began asking the question in March.
- Only 11% described AI’s effect as positive.
- 55% said slowing AI development would be good, while 13% said it would be bad.
The online survey included 1,277 U.S. adults and carries a margin of error of about three percentage points.
Safety beat the race to stay ahead
Respondents were also asked to weigh safety against international competition. Seventy-three percent said the United States should prioritize making AI safe and responsible, while 23% said maintaining the country’s technological lead should come first.
That gap matters because the commercial race is still accelerating. On the same day the poll was published, Alibaba unveiled a faster AI chip and new models as Chinese firms push to narrow the gap with U.S. rivals, according to The Associated Press. The public appears to be saying that speed alone is not a sufficient measure of progress.
Why concern is becoming easier to understand
AI risk can sound abstract when framed around distant scenarios. It becomes more immediate when automated systems handle money, personal information, hiring decisions or access to services.
In a separate report published Tuesday, major banks warned that AI-powered shopping agents could create new openings for scams, fraud and privacy breaches unless payment controls and accountability improve. The concerns described by Reuters show how the debate is moving from experimental chatbots toward tools that can act on a person’s behalf.
The federal government already has a voluntary framework for managing AI risks. The National Institute of Standards and Technology’s AI Risk Management Framework encourages organizations to govern, map, measure and manage potential harms. But voluntary guidance is different from enforceable rules, and the poll suggests many Americans do not believe company-led safeguards are enough.
What the poll does—and does not—say
The findings should not be read as proof that a specific AI disaster is likely. Polls capture public attitudes, and those attitudes can be shaped by news coverage, personal experience and how questions are framed.
Still, this is more than a vague expression of anxiety. Reuters reported that 69% of respondents had followed recent news about efforts to slow AI development and increase oversight. A majority also favored a federal role in setting safety standards.
That creates a widening political tension: voters appear ready for action, while Congress has little time for a quick response. The House is away until November, making near-term federal legislation unlikely.
The next test is whether trust can catch up
AI’s potential benefits remain substantial—from faster scientific research to more capable accessibility tools and productivity gains. Public concern does not necessarily amount to rejection. It is a warning that deployment without visible safeguards may cost the industry the trust it needs to operate at scale.
Companies can answer that warning with clearer incident reporting, independent testing, stronger privacy protections and understandable limits on what autonomous systems are allowed to do. Policymakers, meanwhile, face the harder task of setting durable rules without freezing useful innovation.
The clearest message in the poll is not that Americans want AI to disappear. It is that most do not want the companies building it to be the only ones deciding what counts as safe.
