We’ll Ask the Machine. We Just Don’t Want It Making the Call.

Americans are getting used to artificial intelligence faster than they are getting comfortable with giving it power. That is the clearest tension in a new Rutgers-led national survey: AI is rapidly becoming an ordinary tool, but most people still appear to draw a much harder boundary when the technology shifts from helping with a decision to owning the decision itself.

The latest National AI Opinion Monitor found that 72% of U.S. adults surveyed have used at least one major AI tool, up from 50% in November 2024. More than a quarter, 27%, said they now use AI every day. The technology is also moving deeper into routine life: 68% reported using AI for personal purposes, while 44% of employed adults said they use it at work and 31% said their job requires it.

Those numbers describe a technology that is no longer confined to early adopters. AI is becoming something people encounter because it is useful, because it is convenient, and increasingly because a workplace, school, service or software platform has built it into the expected way of doing things. Familiarity, however, is not the same as surrendering judgment.

When Rutgers researchers asked about high-stakes decisions, support for unsupervised AI authority nearly disappeared. Fewer than one in 10 respondents said they would allow AI to make the final decision without human supervision in hiring, loan approvals, college admissions, parole or determining priority for a medical procedure. Depending on the decision, between 45% and 54% said AI should not be used at all.

The public was more open to a supporting role. Thirty-four percent said AI could assist with hiring decisions without making the final call, while 26% were willing to allow it to assist with parole decisions. Katherine Ognyanova, a professor in the Rutgers School of Communication and Information and a coauthor of the report, summarized the distinction plainly: “Using a tool and trusting it fully are two different things.”

That distinction may be more useful than the familiar question of whether Americans “trust AI.” Trust is often treated as a single scale, as though people begin suspicious, become comfortable through repeated use and eventually accept a machine in progressively more important roles. The Rutgers findings suggest something more categorical may be happening. People can be comfortable asking a system to research, summarize, draft, compare or recommend while still refusing to grant the same system authority over a job, a loan, an education, a release from prison or access to medical care.

In other words, the boundary may not be between people who accept AI and people who reject it. It may be between kinds of permission. Assistance is one permission. Evaluation is another. Final authority is something else entirely.

The demand for disclosure reinforces that separation. Seventy-six percent of respondents said employers should disclose when AI evaluates workers or job applicants. The same share said companies should disclose when AI makes decisions about customers, and 77% said schools should disclose when AI evaluates students. Fewer than 6% disagreed with any of those disclosure requirements.

That is a notable expectation because many consequential automated systems do not arrive with the drama of a robot announcing that it is taking over. They appear as screening tools, scoring systems, recommendation engines, rankings, flags, prioritization models and workflow software. A human may still be present somewhere in the process, but the machine can shape what the human sees first, what gets filtered out and which options appear most defensible. Asking for disclosure is, at minimum, a request to know when that invisible layer is participating in the judgment.

The survey also found broad support for government oversight. Fifty-nine percent said AI should be strictly regulated because of its potential risks, compared with 25% who preferred minimal regulation to encourage innovation. Even where confidence in AI has shifted politically, regulation remained a point of relative agreement: 62% of Democrats and 57% of Republicans favored strict regulation, according to Rutgers.

Concern, meanwhile, has not vanished as use has grown. Half of respondents described themselves as worried about AI, and 48% of daily users said they were worried about it. The share describing themselves as angry rose from 19% in late 2024 to 30% in August 2026. That combination matters because it complicates another easy story about technology adoption: people do not necessarily have to resolve their concerns before a tool becomes part of their life. They may use it and worry about it at the same time.

There is a similar gap between familiarity and understanding. Fifty-nine percent said they understand the basics of how AI works, but respondents averaged 48% correct when judging 12 statements about the technology. Among daily users, 82% said they understood the basics while averaging 49% correct. Routine use, in other words, may produce confidence with an interface much faster than it produces technical knowledge about the system behind it.

None of this means the survey proves why people resist automated final decisions. The questions establish a boundary, not a single explanation for it. Concerns could involve bias, transparency, accountability, the ability to appeal, fear of error, discomfort with opaque systems, or simply the belief that some decisions carry a moral weight that should remain attached to a person. The survey does not reduce those possibilities to one motive.

But it does point toward a more precise public bargain with AI than either enthusiasm or backlash can capture. Americans are increasingly willing to invite machines into the room. They will ask for help finding information, organizing work and producing recommendations. They may even accept AI as one participant in difficult decisions, provided its role is visible and limited. What they are far less willing to do is let the recommendation quietly become the consequence.

That may be the durable line as AI becomes more capable: not whether the machine is allowed to speak, but whether anyone is still responsible for deciding what happens after it does. 

We’ll ask the machine. We just don’t want it making the call.

SOURCE NOTES

• Rutgers University — “Americans Use Artificial Intelligence, but They Don’t Want It in Charge.” Published Sept. 9, 2026. This is the primary source for the 72% adoption figure, 27% daily-use figure, workplace-use findings, high-stakes decision questions, disclosure preferences, regulation attitudes, worry/anger measures, knowledge test, quotes from Katherine Ognyanova, and the 2026 survey methodology
• Rutgers University — “Survey Reveals Concerns and Adoption Trends Around AI’s Rising Influence.” Published Dec. 19, 2024. This is useful background for the first National AI Opinion Monitor wave and provides context for the earlier attitudes against which the 2026 results are compared
• Rutgers University — “Survey Highlights an Emerging Divide Over Artificial Intelligence in the U.S.” Published Feb. 9, 2025. This covers the earlier National AI Opinion Monitor work on AI trust, use and knowledge, based on the late-2024 survey of nearly 4,800 respondents

Source and methodology: Rutgers University, “Americans Use Artificial Intelligence, but They Don’t Want It in Charge,” Sept. 9, 2026. The National AI Opinion Monitor surveyed 5,154 U.S. adults online between July 5 and Aug. 1, 2026. PureSpectrum used a nonprobability sample with representative quotas for gender, age, race, ethnicity and region; researchers reweighted the data to match the U.S. population across multiple demographic and internet-use measures. Official Rutgers release