Anthropic measures the AI trust gap

Anthropic published the first results of Anthropic Public Record on June 12, 2026, based on a nationally representative online survey of 51,993 Americans conducted in November and December 2025. The company says it plans to repeat the series to track how public attitudes toward AI change over time. The sharpest number is not a model benchmark, but a governance signal: only 15% of respondents said they trust AI companies to make decisions about how the technology is developed and used.

The survey shows a fairly stable mix of hopes and fears. On the hopeful side, 48% of Americans ranked curing diseases such as cancer or Alzheimer’s among their top three hopes for AI. Helping people with disabilities followed at 36%, while technological progress and making life easier in general were tied at 23%. On the fear side, job loss was the leading concern in every state, with 64% of respondents saying they were worried. Cognitive dependency, meaning the risk of handing too much thinking over to machines, followed at 56%, ahead of misinformation at 52%.

The most useful finding may be the lack of a simple partisan split. Anthropic says support for government involvement in AI reached 71% nationally, including 79% among Democrats, 68% among Republicans, and 69% among independents. Respondents most wanted action on privacy, child safety, and liability for harm. In other words, the survey does not describe an anti-AI public. It describes a public that sees potential benefits, but wants rules, remedies, and external guardrails rather than leaving the boundaries entirely to labs and vendors.

The method matters when reading the results. Anthropic says the survey was weighted to US Census benchmarks and has a national sampling error of plus or minus 0.6 percentage points. The answers are self-reported and reflect late 2025 attitudes, not observed behavior in workplaces or schools. Still, the sample is large enough to give a useful snapshot, especially on the difference between heavy users and non-users. People who use AI every day at work were less worried about job displacement than people who never use it, 54% versus 70%.

For AI companies, the practical consequence is plain: adoption alone will not create trust. Even integrated users, defined as people who use AI daily for both work and personal life, supported government involvement at roughly the same level as the broader public. That makes it harder to frame regulation as a demand coming only from skeptics or late adopters. In this picture, the public wants medical, practical, and productivity gains from AI, but does not want providers to be the only institutions deciding the limits.