84% of Britons want humans to review major AI decisions

The UK AI control survey found that Britons want help from AI, but with a human holding the final say. Here is how the main answers broke down:

• 84% would grant AI moderate, low or no access to their lives.
• 34% would not be comfortable with AI controlling anything at all.
• 27% are happiest when AI gets things ready and then waits for their approval.
• 23% are fine with AI running routine jobs, as long as they keep sign-off on the important ones.
• 9% say they have no worries about AI whatsoever.

The headline number is worth reading carefully. The 84% is not a vote for any one rule. It groups everyone who wants limited or zero AI access, from people who would let an assistant sort their inbox to people who would not let it near their calendar. What unites them is a clear preference for staying in the loop rather than letting software act on its own.

The survey separates two kinds of worry, and Britons treat them very differently. Science fiction fears are a minority view: just 19% describe AI as a threat to humanity.

The more everyday worry is about blame. Forty percent believe that when an AI goes rogue, the company that made it should be held responsible. Another 14% are concerned that an AI tool could land them in legal trouble, for example by committing fraud or some other illegal act in their name. That second worry is sharper among the youngest adults: 21% of 18 to 24 year olds share it.

In other words, people are less afraid of a superintelligence taking over than of a shopping bot or email assistant doing something stupid with their account, and of being left holding the bill.

Age is the biggest dividing line in the results. Among people over 55, 47% said they would refuse to give an AI agent any control over their lives at all, well above the 34% average.

The 25 to 34 age group sits at the other end. One in five people in that bracket, 20%, say they are comfortable giving AI agents a high level of control. That is still a minority, but it shows where early adopters of agent features are likely to come from.

Younger adults are not simply more trusting, though. The 18 to 24 group is the one most worried about legal trouble, which suggests heavy users are also the people most aware of how these tools can misfire.

The short answer, under current UK law, is usually you. Jonathan Armstrong, a partner at the law firm Punter Southall, pushed back on the idea that AI is a lawless frontier, saying in reaction to the survey that legally, the Wild West view of AI “just isn’t the case.”

His point is that the person using the tool generally owns the results. Armstrong said that in most cases you are responsible for the AI you use and what it does, sometimes even when its behavior surprised you. That sits awkwardly with the 40% who expect the AI’s maker to take the blame.

For now, that gap between what people expect and what the law says is the real story here. Until rules change, a mistake made by an agent acting under your login is likely to be treated as your mistake.

The Censuswide numbers line up with other recent polling that shows British caution growing even as use rises. A September 2026 YouGov survey found 79% of Britons have little or no confidence that AI companies will develop the technology responsibly, and 73% think those companies face too little regulation. The same poll found 52% expect AI’s overall impact to be negative, up from 34% in 2024.

Official figures point the same way. In June 2026, 81% of adults told the Office for National Statistics they fear AI is making fake news harder to spot, and a large majority were also uneasy about their personal data being used without permission. Only 13% of adults thought AI’s benefits outweighed its risks, against 38% who saw more risk than benefit.

Yet use keeps growing. A UK government engagement survey for 2025/26 found 59% of adults had used some form of generative AI in the previous three months. Britons are using these tools, just not handing them the keys.

If you are one of the many people who want AI to help but not to act alone, you can set things up that way today. Most assistants and agent features let you choose how much they can do without asking.

• Turn on confirmations. Where an assistant can send email, make purchases or book things, require it to ask before every action.
• Limit what it can reach. Connect only the accounts it needs. An AI that can read your calendar does not also need your bank login.
• Keep payment details out. Avoid storing card details in agent tools unless you are happy to review every charge.
• Check the terms. Since you are likely to be responsible for what the tool does, read what the provider says about errors and refunds before you rely on it.

Those settings match what the 27% and 23% groups in the survey said they want: AI that prepares the work, and a person who presses the button.