Agentic AI liability: Fraud and liability at scale with Robbie MacDiarmid
Today, we are digging into agentic AI liability and what happens when AI starts doing more than just recommending products. Because once an agent can search, decide, interact with a merchant, and complete a purchase, fraud and payments stop being background functions. They become the whole story.
I sat down with Robbie MacDiarmid to talk about where this is all going, how agentic commerce is being built, and why so many teams are rushing toward the customer experience side of AI without really thinking through the fraud, payments, and liability side of the equation. And yeah, that is a problem.
At first glance, this can sound like a future-state innovation conversation. But when you look closer, it is really a fraud operations conversation. It is about account takeover through AI agents, chargebacks from agentic AI, first-party fraud in agentic commerce, and the unresolved question sitting underneath all of it: who owns liability when an autonomous agent makes the purchase, the payment goes through, and the customer later says they did not authorize it?
That question matters. A lot.
Because we have seen this playbook before. New payment behavior shows up, the user experience gets the attention, and fraud teams get pulled in later to clean up the parts nobody mapped out at the start.
Here is what agentic AI liability means in practice:
- Autonomous agents can introduce new fraud risk across search, merchant interaction, and payment
- AI shopping agent payments may create confusion around authentication, authorization, and intent
- Liability for AI-initiated purchases is still not clearly defined across the ecosystem
- Fraud controls for agentic payments need to be designed before these flows scale
- Merchants, issuers, and platforms all need a clearer trust framework for AI commerce
What you’ll hear in this episode
- How Robbie breaks agentic commerce into search, merchant interaction, and payment
- Why fraud and payments are too often treated like an afterthought in AI-powered commerce
- Where agentic commerce fraud risk is likely to show up first
- Why account takeover through AI agents and first-party fraud in agentic commerce are both real concerns
- What fraud teams should be asking now about payment responsibility for autonomous agents
You should listen to this episode if you
- Work in fraud, payments, or trust and safety and want to understand agentic AI liability before it becomes a bigger issue
- Are thinking about AI shopping agent payments and need to pressure test fraud and liability assumptions
- Support ecommerce fraud and agentic checkout flows inside a merchant, fintech, or payments company
- Want to understand how chargebacks from agentic AI may reshape online commerce liability shift conversations
- Are trying to build a fraud strategy for autonomous transactions without waiting for the losses to force the conversation
If you liked this episode, be sure to subscribe and review the podcast on iTunes, Spotify, YouTube, or wherever you listen to podcasts. It really helps with getting the word out.
Episode notes & key takeaways
Why agentic AI liability is not just a payments issue
Let’s break this down.
A lot of the current conversation around agentic commerce is focused on convenience. Faster discovery. Less friction. More automation. The idea is that an AI agent can handle parts of the shopping journey for the customer, from finding the product to helping complete the purchase.
But here is what is actually happening.
The moment that agent starts taking action on behalf of a person, fraud and liability move to the center. Not the edge. Because now we are not just talking about a shopper browsing with AI assistance. We are talking about AI-powered purchase authorization, transaction intent, payment authentication for AI agents, and whether the person behind the wallet actually meant for that transaction to happen.
That is where the real risk comes in.
If the industry does not define agentic AI liability clearly, every disputed transaction becomes a fight over what counts as consent, what counts as authentication, and who is supposed to absorb the loss.
Here are the pressure points fraud teams should already be watching:
- Whether the customer explicitly authorized the AI agent to make a purchase
- How merchants distinguish human from AI buyer behavior at checkout
- What proof exists if a customer later disputes the transaction
- Whether current fraud controls can handle autonomous agent transaction risk at scale
How fraud shows up in agentic commerce
This might sound like a new category. In some ways, it is. But the underlying abuse patterns are pretty familiar.
We have seen versions of this before with digital wallets, delegated access, saved credentials, and account takeover. Agentic commerce just adds another layer between the person and the transaction. And that layer can make already hard fraud questions even harder.
So what does that actually look like?
It can look like a customer claiming they do not recognize a purchase made by an AI assistant. It can look like criminals using stolen credentials to connect their own agent to a legitimate account. It can look like first-party fraud in agentic commerce, where a cardholder benefits from the purchase and then disputes it anyway because the purchase path feels less direct. And it can look like merchants accepting agent-driven checkouts without knowing whether their existing authentication stack can support that model.
Not exactly subtle.
The key thing to understand is that agentic commerce fraud risk does not need to invent entirely new attack paths to cause damage. It just needs to make existing attack paths more scalable, more confusing, and harder to assign responsibility for.
That usually does not end well.
The liability question nobody has solved yet
This is where things get interesting.
The biggest unresolved issue in this entire conversation is liability for AI-initiated purchases. Because once an autonomous system is acting on a consumer’s behalf, the old assumptions start to break down.
If the AI gets it wrong, who owns that?
If the account was taken over, who proves it?
If the customer claims they never approved the purchase, what standard of proof applies?
And if the transaction technically followed all the current rules, but still resulted in an unauthorized purchase, who takes the loss?
Right now, there is no clean answer.
And that matters because merchants, issuers, networks, and platforms all have different incentives. Everyone wants seamless commerce. Fewer people want ambiguous liability.
A workable trust framework for AI commerce will probably need to answer a few very practical questions:
- How do we verify that the agent is acting for the actual account owner
- What does know your agent verification need to look like in real life
- How should payment responsibility for autonomous agents be allocated
- What evidence should be retained for disputes, fraud reviews, and chargebacks
Why fraud teams need to get involved early
I have been seeing this pattern more and more lately.
A new technology starts gaining traction. Product teams get excited. The conversation centers on adoption, efficiency, and customer experience. Then later someone asks the fraud team how to make it safe after the model is already live.
Yeah. That is backwards.
Fraud teams should be in these discussions now, especially when it comes to merchant interaction with AI agents, digital wallet agent risk, and ecommerce fraud and agentic checkout design. Because if fraud is treated as a final-stage control, the whole system is more likely to inherit weak assumptions around identity, authorization, and dispute rights.
This is one of those moments where asking better questions early can prevent a lot of operational pain later.
Questions like:
- How will we authenticate an agent at checkout
- What counts as valid authorization for an AI-initiated purchase
- How do we investigate account takeover through AI agents
- What data will help us separate legitimate use from fraud or abuse
- Who is responsible when the customer, the merchant, and the AI all tell a slightly different story
What fraud fighters should be watching next
If you work in fraud, this probably sounds familiar.
The technology changes. The core questions usually do not.
Who is the real user?
Was the transaction actually authorized?
What evidence do we have?
Who takes the loss?
Agentic AI just adds a new layer of complexity to all four.
That is why this conversation with Robbie matters. It is not just about the future of shopping. It is about the next wave of fraud risk, the next set of chargeback questions, and the next liability fights that are going to show up once these models move from experimentation to scale.
So the takeaway here is pretty simple.
Do not wait for the losses to teach the lesson. Start defining your fraud controls for agentic payments now. Start mapping autonomous agent transaction risk now. And start pushing for a clearer trust and liability model before AI commerce gets far enough along that everyone is arguing after the fact.
Because once that happens, fraud teams will still be the ones asked to make sense of it.
