AI Hallucinations & Mastercard’s New Scam Thresholds

In this episode of Fraudology, Karisse Hendrick comes to you from a beachside work-cation in Florida to deliver an essential debrief on the latest shifts in the e-commerce fraud landscape. Fresh off the Accertify Global Customer Summit, Karisse shares key strategic takeaways on why cybersecurity and fraud teams must break down operational silos as fraud signals increasingly move up-funnel.

The conversation takes a critical look at the limitations of relying on Large Language Models (LLMs) in risk management. Highlighting a recent blunder where a Top 4 consultancy published a 44-page fraud report riddled with completely fabricated citations and footnotes, Karisse and Dr. Nicola Harding explain why "domain expertise" cannot be automated. Because true fraud insights are kept proprietary to protect them from criminals, open-source AI tools are inherently prone to "hallucinating" facts.

We also break down Mastercard's newly announced Scam Merchant Dashboard, which officially goes into effect on July 24th, 2026. This aggressive program places a heavy burden on e-commerce merchants and their acquirers through a multi-trigger framework designed to shut down predatory accounts.

Key pillars of Mastercard’s new program include:

• The Authorization Performance Breakdown: A sudden drop in approval rates—such as a 50 percentage point decline or falling below a 30% overall threshold within a 72-hour window—will immediately trigger an investigation.
• The New Merchant 5% "Math": For accounts open less than six months, Mastercard is introducing a brand-new metric: combining refunds and chargebacks divided by overall sales. Crossing a 5% threshold over a rolling 30-day period (with at least 500 transactions) risks immediate account review.
• The 72-Hour Termination Clock: Once flagged by issuer complaints or network alerts, acquirers have a strict 72-hour window to either prove the merchant's legitimacy or completely terminate their ability to accept Mastercard.