Welcome back to Fraudology.
I’m coming to you from a beachside work-cation in Florida for this one, right after the Accertify Global Customer Summit. And yes, the setting was beautiful.
In this episode, I’m joined by Dr. Nicola Harding to talk through two things that may seem separate at first: AI hallucinations in fraud and Mastercard’s new scam merchant monitoring program.
When you look closer, they are connected by the same bigger issue.
Fraud teams are being asked to make faster, more complex decisions in environments where the signals are moving earlier, the liability is shifting, and the tools we use to interpret risk are not always as reliable as they sound.
That matters.
On one side, we have LLMs being used in risk management, fraud research, and operational decision-making. And as we discuss in this episode, that gets risky fast when AI starts producing confident answers from incomplete or open-source information. Especially in fraud, where the most useful knowledge is often proprietary for a reason.
On the other side, Mastercard’s scam merchant monitoring program creates a much more aggressive framework for identifying and investigating merchants that may be tied to scams. The new thresholds, refund and chargeback monitoring, and 72-hour merchant investigation window put real pressure on merchants, acquirers, and payment teams.
So this episode is really about accountability.
Who owns the decision?
Who validates the signal?
Who understands the context?
And what happens when either a model or a merchant program gets treated like it can operate without the right human expertise around it?
What you’ll hear in this episode:
- What Mastercard’s scam merchant monitoring program means for merchants and acquirers
- Why the Mastercard scam merchant dashboard matters for e-commerce fraud teams
- How refund and chargeback monitoring may affect new merchant risk reviews
- Why the 72-hour merchant investigation window creates operational urgency
- How AI hallucinations in fraud can distort risk analysis and decision-making
- Why LLM hallucinations are especially risky when fraud knowledge is proprietary
- Why fraud and cybersecurity teams need to break down silos as signals move earlier in the attack path
You should listen to this episode if you:
- Work in fraud operations, merchant risk, payments, or e-commerce fraud
- Are responsible for Mastercard chargeback thresholds or scam merchant investigation workflows
- Need to understand how the Mastercard scam merchant monitoring program may affect your team
- Are evaluating LLMs in fraud risk management or AI-generated fraud research
- Care about domain expertise, fraud and cybersecurity alignment, and operational readiness
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:
This episode sits at the intersection of two very different but very important fraud problems.
The first is AI hallucination risk. Not the funny kind where a chatbot makes up something harmless and everyone moves on. I mean the kind where an AI tool produces a confident fraud analysis, citation, recommendation, or risk interpretation that is not actually grounded in reality.
That is a problem.
Fraud teams do not operate in a world where all the useful information is public. A lot of the best fraud intelligence lives inside internal systems, proprietary rules, investigations, chargeback patterns, merchant histories, cybersecurity data, and operational experience. So when an LLM tries to reason from open-source data alone, it can miss the exact context that matters most.
The second issue is Mastercard’s scam merchant monitoring program, which puts more pressure on merchants and acquirers to identify, investigate, and act on scam-related merchant activity quickly. This is not just a policy update. It is an operational readiness issue.
And that is where the two themes connect.
Fraud teams are being asked to move faster while the risk signals get more complex. That means the companies that do well here will not be the ones that blindly trust every dashboard, every model output, or every surface-level metric.
They will be the ones that understand the signals, validate the data, and bring the right domain expertise into the decision.
Why Mastercard scam merchant monitoring changes the pressure on acquirers
Mastercard’s scam merchant monitoring program is designed to identify merchants that may be connected to scams, deceptive activity, or harmful selling practices. That means the focus is not just traditional card-not-present fraud. It is merchant behavior, refund patterns, chargebacks, issuer complaints, authorization performance, and whether a merchant looks like it may be creating risk for the network.
For acquirers, that changes the posture.
It is not enough to wait until the damage is obvious. Acquirers need to be able to investigate quickly, understand merchant risk signals, and make a call on whether a merchant is legitimate or needs to be terminated.
That is a very different kind of pressure.
- Mastercard scam merchant monitoring increases the need for faster merchant risk review
- Acquirers may need stronger workflows for scam merchant investigation
- Merchant risk teams should watch refund, chargeback, and authorization patterns together
- The 72-hour merchant investigation window makes operational readiness critical
Why refund and chargeback monitoring matters for new merchants
One of the most important pieces of this program is the way it looks at refunds and chargebacks together for newer merchant accounts.
Scam merchants do not always show up through one clean signal. Sometimes the pattern is a mix of complaints, refunds, chargebacks, low authorization performance, and behavior that looks just legitimate enough to keep operating.
At first glance, a refund might look like customer service.
But when you dig in, refunds can also be a signal that something else is happening. Especially when they are paired with complaints, chargeback activity, or sudden approval rate changes.
For fraud teams, the takeaway is simple: do not look at these signals in isolation.
- Refund and chargeback monitoring can reveal scam patterns earlier
- New merchant accounts may need closer review during the first six months
- Mastercard scam thresholds create a stronger incentive to monitor risk continuously
- E-commerce fraud teams should connect merchant behavior to customer complaint signals
Why AI hallucinations are dangerous in fraud risk management
AI hallucinations in fraud are risky because they can create confidence where there should be caution.
A model can summarize. It can draft. It can organize information. It can even help teams move faster.
But if the source data is incomplete, wrong, or missing the proprietary context that fraud teams rely on, the output can fall apart very quickly.
This is where domain expertise matters. A fraud professional can look at a polished AI-generated statement and ask, “Wait, does that actually make sense?” A model does not always know when it is outside its lane.
And in fraud risk management, that distinction matters.
- LLM hallucinations can distort fraud analysis and operational recommendations
- Open-source AI tools may miss proprietary fraud patterns
- Domain expertise helps teams validate whether an AI output is usable
- AI should support fraud operations, not replace human judgment
Why fraud and cybersecurity silos are becoming a bigger liability
One of the bigger themes coming out of the Accertify Global Customer Summit was that fraud signals are moving up-funnel.
That means the first signs of risk may not show up at the transaction anymore. They may show up earlier through account activity, device behavior, phishing, malware, credential abuse, or other cybersecurity signals.
So if fraud and cybersecurity teams are still operating in separate lanes, that creates a gap.
And criminals tend to like gaps.
Fraud teams need visibility into the signals that happen before payment. Cybersecurity teams need to understand how those signals eventually turn into e-commerce fraud, payment fraud, merchant risk, or chargebacks.
The more connected those teams are, the earlier they can see the pattern.
- Fraud and cybersecurity silos make it harder to detect risk early
- Up-funnel signals can help teams understand fraud before the transaction
- Payment fraud prevention works better when teams share context
- Cross-functional visibility helps reduce blind spots in fraud operations
Why domain expertise still has to anchor the decision
This episode keeps coming back to one point: tools are useful, but expertise is still what turns information into judgment.
That applies to AI hallucinations. It applies to Mastercard scam merchant monitoring. It applies to merchant risk. It applies to fraud and cybersecurity alignment.
A dashboard can flag a threshold.
A model can summarize a pattern.
A report can identify a risk.
But someone still has to understand what it means.
That is where experienced fraud professionals matter most. They know when a merchant pattern looks wrong. They know when a policy threshold needs operational support. They know when an AI answer sounds too clean. They know when different signals are pointing to the same underlying problem.
The cost of getting this wrong is not just a bad report or a messy workflow. It can be losses, liability, terminated merchants, missed scam networks, or decisions made from information that was never properly validated.
Final takeaway
The Mastercard scam merchant monitoring program is another reminder that fraud prevention is becoming more connected, more time-sensitive, and more operationally demanding.
At the same time, AI hallucinations are a reminder that faster information is not always better information.
So the real takeaway is not just “watch the thresholds” or “be careful with AI.”
It is this: fraud teams need stronger context.
Context across merchant risk.
Context across fraud and cybersecurity.
Context across refunds, chargebacks, complaints, and authorization behavior.
Context across AI outputs and the proprietary knowledge that models do not automatically have.
Because in fraud, the tools can help you see more.
Connect with Karisse Hendrick | LinkedIn
Host of the Fraudology Podcast
Award-Winning Cyberfraud Expert
Ecommerce Fraud Prevention Consultant
Startup Advisor, Keynote Speaker, and
Consultant to Fortune 500 merchants
