A new survey of 52,000 U.S. adults, commissioned by Anthropic, the AI company behind the Claude chatbot, found 64% fear job loss and 56% fear losing independent thought to AI, but those who actually use the technology report meaningfully lower fear — 54% versus 70% among non…
The clearest signal from a new survey of 52,000 Americans is that they do not want one thing from AI. They want two opposite things at once. The Decoder reports that 48% of U.S. adults list curing serious diseases like cancer and Alzheimer's among their top three hopes for the technology, while 64% fear AI will cost them their job and 56% fear losing the ability to think for themselves. The hope and the fear are two sides of one story about what Americans want AI to do and where they want it stopped.
The numbers come from a YouGov online poll fielded between November and December 2025, covering 51,993 Americans aged 16 and older across all 50 states, the District of Columbia, and Puerto Rico, as summarized in The Decoder's writeup of the survey. Anthropic, the AI company behind the Claude chatbot, published the results on June 12, 2026, as the first installment of a recurring series it calls the "Anthropic Public Record," explicitly aimed at the general public rather than existing Claude users. That provenance matters: this is a frontier model vendor surveying attitudes toward its own industry, and the framing reflects that origin.
What the survey shows is that the fear numbers are not abstract. Job-loss fear ranked first in every state, running from 71% in Iowa to 57% in Mississippi, and it rose with education level, peaking among the groups most exposed to the office work AI tools are already reaching. Americans with postgraduate degrees were nearly 10 percentage points more worried about job loss than those with a high school education or less. The 56% who fear losing independent thinking is a different kind of alarm, naming cognitive dependence rather than economic displacement. It suggests the public is reading AI as a force that can quietly reshape how decisions get made, not just who gets to make them.
Against that, the experience finding is the most consequential detail in the data. Respondents who reported actually using AI tools at work every day were notably less worried about job loss than those who did not use AI at all: 54% versus 70%. The pattern lines up with years of consumer-tech research showing that direct use tends to compress early fear of new tools once the use case is concrete. Translated into policy and product terms, the survey is not measuring a fixed public mood. It is measuring a public mood that moves the moment people get hands on the technology.
The most uncomfortable number for AI companies is the trust number. Only 15% of respondents said they trust AI companies to make the right calls about how the technology is deployed. That is a separate axis from the fear numbers: it is a verdict on the institutions doing the building, not on the technology itself. Combined with widespread skepticism about workplace AI deployment, the trust gap suggests that the public is not simply anti-AI. It is willing to use the technology and unwilling to leave the deployment decisions to the companies making it.
That distinction is what the experience finding actually sharpens. Americans are not asking AI companies to slow down. They are asking for AI products they can use on their own terms, in their own workplaces, without ceding judgment about how the technology is rolled out. The survey frames adoption as a behavior problem, not a sentiment problem, and the answer it points to is structural: transparent deployment, clear limits, and a voice in decisions for the people whose work the technology is entering.
The numbers come with limits. They are self-reported attitudes from a single wave of polling, and Anthropic commissioned the work and published the writeup, so the editorial voice belongs to a model vendor. The exact figures, question wording, and cross-tabs have not been independently verified against Anthropic's primary publication. What the data does support, plainly, is that AI is arriving in a public that wants its benefits and is wary of the people steering it, and that the gulf between those two feelings shrinks the moment more Americans get to use the tools themselves.