Anthropic Surveyed 52K Americans - Just 15% Trust AI
Anthropic's first Public Record survey of 51,993 Americans finds 64% fear job displacement, only 15% trust AI companies, and 70%+ support government regulation - with rare bipartisan consensus.
7 min read
An AI company has just published survey data showing that almost nobody trusts AI companies.
That's the short version of Anthropic's first Public Record, released June 12 - a nationally representative survey of 51,993 Americans conducted in November and December 2025. The sample covered every state, with 232 to 1,902 respondents per state and margins of error between ±2.6 and ±9.1 percentage points. It's a serious research exercise, not a PR exercise, and the numbers are uncomfortable for the industry.
TL;DR
- 51,993 Americans surveyed across all 50 states (Nov-Dec 2025), results published June 12
- 64% fear job displacement from AI - the single most common fear in every state
- 15% trust AI companies to make key development decisions - lowest of any institution tested
- 71% want government involved in AI regulation, with bipartisan majority support
- Daily AI users fear job loss notably less (54%) than non-users (70%)
What Americans Hope and Fear
Before getting to the trust numbers, the survey asks a simpler question: what do you want AI to do, and what are you afraid it'll do?
| Hopes (top 3 ranked) | Share | Fears | Share |
|---|---|---|---|
| Cure diseases (cancer, Alzheimer's) | 48% | Job displacement | 64% |
| Help people with disabilities | 36% | Cognitive dependency | 56% |
| Technological progress / easier daily life | 23% | Spread of misinformation | 52% |
The hope side skews altruistic and medical. The fear side is led by economic anxiety. Job loss doesn't just lead - it's the top fear in every single state, ranging from 57% in Mississippi to 71% in Iowa. There's no regional exception. The response doesn't segment neatly by partisan identity or geography.
Job displacement is the top AI concern in every US state, regardless of region or political leaning.
Source: unsplash.com
The Trust Number in Context
Who Americans Actually Trust
The headline finding is that only 15% of Americans trust AI companies to make decisions about how AI is developed. That's the lowest score of any institution Anthropic tested. Independent experts scored highest, at 43%. Government agencies and academia sit somewhere in between.
That 15% figure needs context. It isn't a specific verdict on Anthropic, Claude, OpenAI, or any individual lab. It's a general read on the industry, collected from a population that includes many people with limited direct exposure to AI products. But the number matters because AI labs - Anthropic included - regularly claim they're developing AI in the interest of humanity. This survey shows roughly six in seven Americans don't buy that claim.
What Would Change Their Minds
When the survey asks what would best ensure AI develops for the public good, two answers pull away from the rest. Legal liability for harm (47%) and prioritizing safety over growth (44%) score highest. Both are structural demands, not requests for better communication or PR. Americans aren't asking to be better informed about AI development. They're asking for external accountability mechanisms that don't currently exist at the federal level.
Job Loss Fear Shows No Signs of Fading
Anthropic's own job risk exposure research published in March found computer programmers and knowledge workers at 75% AI usage exposure - actual measured exposure, not theoretical risk. That work, combined with evidence from layoff data showing 55,000 workers cited AI as a factor in their job cuts last year, gives the 64% fear number a credible foundation. People aren't afraid of a vague future. They're watching it happen to colleagues.
The Power User Exception
The survey surfaces an interesting gap between frequent users and everyone else. About 6% of Americans use AI daily for both work and personal tasks - what Anthropic calls "integrated users." They skew young, urban, and college-educated. Among this group, 54% worry about job displacement. Among non-users, that number jumps to 70%. Regular users, having built their own workflows and seen AI augment rather than replace their work, are meaningfully less anxious about it.
More Education, More Fear
The education correlation cuts the other direction. Job loss concern increases with education level and peaks among postgraduate degree holders. This isn't hard to explain: white-collar knowledge work is exactly where language models have the most obvious near-term application. Lawyers, analysts, consultants, and researchers can see what these tools do. They're not unfamiliar with AI's capabilities - they're more familiar with them than most.
Postgraduate degree holders express the highest levels of job displacement concern among education groups surveyed.
Source: unsplash.com
Educators are a separate case. The survey notes they report 2.5 to 3 times higher cognitive atrophy concerns than other groups - the worry that reliance on AI erodes the thinking skills it's meant to support. That anxiety is specific to their professional context, where students using AI to complete work is a concrete daily problem, not an abstraction.
Regulation Gets Rare Bipartisan Support
71% of Americans say the government should be involved in AI development and regulation. The support is bipartisan. In an era where most policy questions split along partisan lines almost automatically, that breadth is remarkable.
What Americans Want Done
When the survey asks which areas to focus on, the answers cluster around data privacy (56%), child safety (52%), and corporate liability for AI-caused harm (49%). These aren't requests for a broad technology regulatory framework. They're specific asks for the government to address concrete failure modes - data misuse, harm to minors, and legal accountability for mistakes.
71% of Americans want government involved in AI regulation. Support is bipartisan, with no regional exception.
State-level variation in regulation support runs from 63% in Hawaii to 81% in DC. Even the floor - 63% - represents a large majority. This isn't a close call in any state.
The bipartisan consensus is worth stacking against the current federal situation. The US has no comprehensive federal AI law. Executive orders on AI have shifted with each administration. In parallel, the EU AI Act continues its complex rollout and state-level action in the US is accelerating. Americans want federal coordination; they're not getting it.
Americans across party lines support government involvement in AI regulation, but federal action remains limited.
Source: unsplash.com
What This Survey Does Not Tell You
A survey of this scale can reveal what people believe. It can't tell you why. The 64% who fear job loss - are they responding to actual observed disruption, to media coverage, or to a general sense of technological anxiety that has accompanied every major shift from steam engines to personal computers? The survey doesn't distinguish.
The 15% trust figure is similarly context-free in one important direction. Anthropic published a separate qualitative study of 80,000 Claude users in March, which found users' top aspiration for AI was more time with family, not pure productivity gains. That's a warmer picture than the general population survey. The difference is selection: Claude users chose to engage with the technology. The 85% of Americans who don't trust AI companies are largely the same people who haven't integrated it into their daily lives.
That doesn't make the distrust invalid. If AI systems are going to affect everyone - workers who don't use Claude, children whose parents never touch ChatGPT, communities where no one opted into anything - then public trust matters regardless of product adoption rates. Consent to use an AI assistant and exposure to AI's economic effects are different things.
Anthropic plans to repeat the Public Record survey regularly and expand it beyond the US. The next waves will track whether trust moves as AI becomes more embedded in daily life. The hypothesis the industry is implicitly betting on is that familiarity breeds acceptance - the same pattern as smartphones, social media, and search. That bet could be right. But the integrated user data in this survey offers a caution: even among daily AI users, 54% still worry about job displacement. Familiarity reduces the fear. It doesn't remove it.
Sources:
