Most people say no, for now. In 2026 a major research firm found three-quarters of online adults in the US, UK and Canada were uncomfortable letting an AI agent complete a purchase and pay on its own, even with preset spending limits. What changes minds is control: approving each payment, clear limits, someone to call when it goes wrong, and someone accountable for the mistake.
A note on the legal and regulatory points here: they summarise public sources as of November 2026 and should not be relied on. Rules for AI agents are still being written, and some of what is described here is proposed or coming soon rather than settled. Check the current position, or ask a qualified professional, before acting on it.
What exactly are people saying no to?
People are mostly saying no to the payment step, not to AI shopping help. The same research firm found many consumers comfortable with an agent recommending and comparing products, and some with it filling a cart. Most drew the line at paying.
That distinction matters. An AI shopping agent, meaning software that can search, choose and buy on your behalf, does several jobs. Each one hands over a little more control:
- Recommend. It suggests options. You decide.
- Compare. It ranks options against your criteria. You decide.
- Prepare. It fills the cart. You review and pay.
- Pay. It completes the purchase without asking.
The first three are advice. The fourth is action. A bad recommendation costs you a minute. A bad purchase costs you money, and possibly a fight to get it back.
The research firm found the few people who would accept autonomous payment mostly meant routine or low-cost purchases, or cases governed by strict preset rules and thresholds.
Why don’t people trust an AI to pay?
People don’t trust an AI to pay because they doubt it will act in their interest, and doubt anyone will fix it when it doesn’t. The numbers are consistent across several 2026 surveys.
A UK survey of 2,080 adults, run in June 2026 for a large payments technology company, found:
- 18% trusted AI to act in their best financial interest.
- 15% trusted AI to handle problems when something goes wrong.
- 17% trusted it to keep personal and payment data secure.
- 60% would stop using an AI shopping agent after a single mistake.
- 70% said purchases made without asking would affect their willingness to use one.
The research firm grouped people’s concerns into four themes: loss of control, errors and liability, distrust of the agent’s decisions, and data security and privacy. A separate seven-country survey of about 8,000 consumers in May 2026, run for another payments company, put payment security (50%), privacy (46%) and wrong purchasing decisions (42%) at the top.
Notice the split. Half of the UK sample trusted AI to find the best price, but fewer than one in five trusted it to act in their interest. The doubt is less about whether it can shop than about whose side it is on, and who fixes its mistakes.
Is anyone saying yes?
Yes, for small, routine purchases and among crypto users. The yes is narrower than the headlines suggest.
The seven-country survey found people expected AI agents to make 15% of their purchases within five years, up from 9% in its previous wave in late 2025. Comfort was highest for small amounts. Large majorities said they would let an agent spend up to $50 on things like cinema tickets or meal delivery. The survey was paid for by a company that sells payment services, and comfort with a hypothetical is not the same as handing over a card.
Crypto users are more open, but split on the key question. A 2025 survey of 2,632 crypto users by a major crypto data platform found 87% would let an AI agent manage at least a tenth of their portfolio. About one in seven would hand over all of it. Yet only 34.5% said AI agents could be trusted with their wallets, against 37.5% who said they could not.
The gap between those two crypto numbers is the whole story in miniature. Letting an agent “manage” part of a portfolio can mean anything from alerts to suggestions. Letting it into the wallet means letting it move money without asking.
Our reading: people are open to AI with money in principle, and wary in the specific moment of letting it spend.
What would change people’s minds?
Control and accountability would change minds, more than better AI. Asked what would help, people describe safeguards they can see and use, not smarter models.
| What people ask for | What the surveys found |
|---|---|
| Approve every payment | 33% in the seven-country survey wanted to approve each transaction before payment |
| Hard limits | 43% of UK adults trusted AI to follow spending limits, far more than trusted its judgement |
| A human to call | 28% in the seven-country survey wanted human support when problems arise |
| Proof it cannot be tricked | 26% wanted proof the system cannot be hacked or manipulated |
| Someone accountable | 54% of UK adults said the company that built the agent should be accountable for refunds |
The research firm’s list was similar: clear controls such as spending limits and purchase approval, transparency about what the agent did and why, strong protections across the system, and responsive support when things go wrong.
Two things stand out. First, the most requested safeguard, approval of every payment, is the one that makes the agent least autonomous. Second, people want the builder of the agent, not themselves, to carry the risk. Whether that is how responsibility actually falls is a separate and unsettled question.
What could go wrong?
The single mistake. If 60% of people would quit after one error, an agent has almost no margin. A wrong size, a double order or a purchase from a dubious seller can end the relationship.
Purchases you didn’t expect. An agent that buys without asking may be doing exactly what its settings allow, and still surprise you. The gap between “permitted” and “wanted” is where most complaints will come from.
Manipulation. An agent that reads product pages and reviews can be steered by them. A seller that games the agent’s ranking, or hides instructions in a page, can redirect a purchase. This is why proof against manipulation shows up in people’s demands.
Data. A shopping agent needs your payment details, your address and often your browsing history. In the UK survey, 61% said linking an agent to a bank account would affect their willingness to use it, and 54% said the same of tracking their browsing.
Nobody to trust. Perhaps the starkest finding: 59% of UK adults said they would not trust any organisation to manage AI-powered shopping and payments. Banks and regulated payment providers were the most trusted, chosen by 20%.
The trust takeaway
Saying no to an AI that pays on its own is a reasonable position today, not a technophobic one. The tools to make it safer exist in part: spending limits, approval prompts and agent identity checks. The accountability does not yet.
If you do let an AI buy something for you, check:
- Does it ask before paying? Start with approval on every purchase, then loosen it only for small, routine buys.
- What is the cap? A per-purchase and monthly limit you would be comfortable losing.
- Can you see what it did and why? A clear log of each purchase and the reason.
- Who refunds a mistake? Read the terms before you need them.
- What data does it hold? Payment details, address, browsing history, and whether you can delete them.