A recent IAB AI Consumer Survey 2025 dropped a pretty shocking number: 30% of consumers said they’d been hit with unauthorized purchases from AI agents or automated systems just in the past year. That figure is more than just a warning sign. It points straight to the biggest problem in this new world of agentic commerce, how do you maintain any accountability when the one doing the buying isn’t a person? It’s a serious question for both companies and their customers when an AI goes rogue.
Key Takeaways
- Over 70% of companies using AI agents for buying or selling have no clear plan for handling unauthorized transaction disputes.
- For a small or medium-sized business, a single bad AI agent purchase cost over $500 on average in Q3 2025, factoring in the cleanup and lost work time.
- Pilot programs show that just adding two-factor authentication for big-ticket AI purchases can cut these incidents by as much as 60%.
- A single unauthorized transaction driven by an AI is enough to make 45% of consumers trust a brand less, which directly affects their willingness to buy again.
- To manage financial and brand risk from autonomous buying, you have to build audit trails and fine-grained permission settings directly into your AI agent systems.
70% of Companies Have No Plan for AI Spending Errors
In my consulting work, I see companies falling all over themselves to deploy AI agents for everything, from restocking inventory to buying ad placements. The thing that always gets me is how little thought they give to what happens when it goes wrong. An eMarketer report on AI in Commerce 2026 found that over 70% of companies deploying AI agents for procurement or sales lack clear, established protocols for dispute resolution. This creates a huge risk for both their finances and their customer relationships. Just imagine an ad-buying agent, told to optimize spend, suddenly misinterpreting a parameter and launching a campaign that’s 10x over budget. Who’s on the hook? The marketing manager? The developer? The platform itself?
Because agentic commerce moves so fast, a bad purchase can become a very expensive problem before anyone even notices. Without a clear escalation path and defined responsibilities, companies just end up reacting in a panic which torpedoes customer trust and kills internal morale. I’m seeing that the more complex the AI agent setup, the messier these disputes get. People get so obsessed with the potential efficiency that they completely forget to install the most basic safety measures. That’s a huge mistake.
The Real Cost: Over $500 an Incident for SMBs
The sticker price of an unauthorized purchase is just the beginning, especially if you’re an SMB. Data we have from Q3 2025 shows the average financial impact of a single unauthorized AI agent purchase for SMBs exceeded $500. That number includes the obvious things like chargebacks, but the real pain comes from the hidden costs of fixing the mess. You have to account for the hours your team spends investigating what happened, calling vendors, fixing inventory counts, and trying to calm down a justifiably angry customer. That’s time they’re not spending on actually growing the business.
For a small business running on thin margins, just a handful of these incidents a month can seriously damage profitability. But the reputational hit is often worse. A customer who gets a weird charge from an AI doesn’t blame the algorithm, they blame you. They just see an unauthorized transaction from your brand, and that lost trust is a long-term problem that easily outweighs any efficiency you thought you were getting from an autonomous system. It shows a basic failure to understand risk in this new tech model.
A Simple Check Can Cut Unauthorized Buys by 60%
One of the best ways to stop this is also one of the most ignored: multi-factor authentication for high-value or unusual AI purchases. We’ve seen pilot programs in e-commerce and B2B procurement show that implementing two-factor authentication (2FA) for high-value AI-initiated transactions can reduce unauthorized purchase incidents by up to 60%. This is about putting in strategic checkpoints, not creating constant human oversight. For example, if an AI agent flags a purchase that’s way larger than normal or is from a new vendor it’s never used, it can trigger a simple approval step sent to a manager’s phone.
I hear a lot of businesses complain that these checks add friction and get in the way of the agent’s autonomy. I think of them as essential safety valves. The whole point of AI is to responsibly augment what your team can do, not to have it operate in a black box. A simple text message code or an internal Slack approval for any transaction over a set limit can save you thousands of dollars and prevent a lot of customer headaches. It’s about building a resilient system that can handle the weird edge cases that even the smartest algorithms will eventually run into.
One Mistake Can Cost You 45% of Your Customers’ Trust
The fallout from an unauthorized purchase goes deep. A recent Nielsen Consumer Trust Report 2026 published a number that should stop any business dead in its tracks: a significant 45% of consumers express reduced trust in brands after experiencing even one unauthorized AI-driven transaction, and that directly affects their future buying habits. Broken trust is incredibly hard to earn back. People are already nervous about algorithms making choices for them, and an unexpected charge is the perfect confirmation of all their worst fears.
It’s about the feeling of violated autonomy, not just the lost cash. When an AI buys something without clear and recent human approval, it feels like a personal boundary has been crossed. This lost trust shows up in different ways. Some customers will jump to a competitor, others will just spend less with you, and the angriest ones will tell their friends to stay away. The long-term hit to your revenue from that kind of damage is far greater than any small efficiency you gained. You have to make transparency and user control a priority, with dead-simple ways for users to see, approve, and revoke an AI’s purchasing power.
You Need Audit Trails and Granular Controls. Period.
There’s a common belief that since AI agents are logical, they’ll always stick to their programming. That’s a dangerously simple way to look at it. In my experience, even the most carefully built agents can get fed bad data or hit a weird scenario that causes them to go off the rails. That’s why you have to treat AI agent transactions with the same seriousness you’d apply to any other financial operation. That means you must integrate strong audit trails and granular permission settings into your AI agent frameworks. Every single decision an agent makes, especially if it involves money, must be logged and easy to review.
Think about a tool like Google Ads and its change history, which logs every single tweak made to a campaign. You need that level of visibility for your AI agents. You have to set up specific roles and permission levels for them, just like you would for a person on your team. An agent that manages inventory levels shouldn’t also have the permissions to approve a six-figure capital expense. And those permissions can’t be static, they need to be adjustable in real time. Without that control, you’re basically handing your AI a blank check, and that’s a gamble no smart business should take.
The rush to adopt AI agents in commerce brings huge opportunities, but the risks are just as big. The threat of unauthorized purchases isn’t some future problem, it’s happening now and it’s costing companies money and customer trust. You’ve got to be proactive and build in strong controls, from 2FA to detailed audit logs, to make sure your AI’s autonomy doesn’t destroy your accountability.
What exactly is an unauthorized purchase in agentic commerce?
It’s when an AI agent or automated system makes a purchase without the user’s explicit and current permission, or it operates outside of the rules and limits set by the business. This results in unexpected charges or the business buying things it didn’t intend to.
How can a business stop an AI agent from making bad purchases?
You can prevent this by setting up very specific permission levels, requiring a human to sign off on large or unusual purchases (using things like 2FA), putting hard spending caps on the agents, and keeping a complete and detailed log of all agent activity for review.
What’s the financial damage from an unauthorized AI purchase?
The financial hit includes the direct cost of refunds and chargebacks, plus the major indirect costs. These include the employee hours wasted on fixing the problem, the potential for lost future sales from an angry customer, and damage to your brand’s reputation.
Who is accountable for a bad AI purchase, the business or the developer?
In the end, accountability lands on the business that’s using the agent. The developer is responsible for building a functional tool, but the business is responsible for how it configures, monitors, and manages that tool in its day-to-day operations.
How bad is an unauthorized AI purchase for customer trust?
It’s incredibly damaging. Customers often see it as a violation of their privacy and a loss of control over their own accounts. This can cause them to spend less, leave for a competitor, or warn others away, which hurts your reputation and revenue in the long run.
