Life in the fast lane: Outpace fraud with the power of AI

Fraud is no longer a background concern—it’s a front-and-center threat to credit unions everywhere. That’s the reality we’re facing.

“We always hear AI is used for good—and used for bad,” said SRM managing director Larry Pruss. “Fraudsters are exploiting AI for various scams. It’s getting more difficult to detect.”

AI technology is no longer exclusive to defenders—it’s now in the hands of fraudsters. We’re seeing new fraud vectors emerge at a pace that traditional systems simply can’t keep up with. AI for fraud prevention is no longer optional—it’s mission-critical.

In 2024 alone, U.S. consumers reported $12.5 billion in fraud losses—a staggering 25% increase over the previous year. And, nearly a third of institutions lost more than $1 million in direct fraud losses, not including the reputational damage and member attrition that often follow.

During a recent roundtable event, “Credit Unions Shift into High-Gear with AI Fraud Detection” Larry and I were joined by Visa SVP, Head of Data and AI Sam Hamilton, where we discussed this hot-button issue and came to a collective conclusion that staying ahead requires a industry-wide, fundamental shift in how we think about fraud detection and prevention.

Agentic AI is changing the game. Unlike traditional models, agentic systems understand context, make decisions, take action and learn from outcomes. Unlike static LLMs, agentic AI augments a knowledge base with real-time learning and decisioning—essential in today’s dynamic environment.

“Agentic capabilities are opening up a lot more things that we can let the computer do . . . taking out some of the cognitive burden from our head into the computer,” Hamilton said.

We’re at a tipping point—similar to the shift from dial-up to broadband or from desktop to mobile. The convergence of bandwidth, GPS and mobile computing created entirely new use cases.

Agentic AI is doing the same for fraud detection. And with quantum computing on the horizon, we’re looking at a future where the pace of innovation will only accelerate.

The democratization of AI means that powerful tools are available to anyone—including those looking to exploit them. Generative models are being used to create deepfakes, impersonations and synthetic identities. Our response must be equally sophisticated.

“We really have to have AI to fight AI,” Hamilton said. “We can't fight that with the previous version of AI or methodology. We need to constantly bring in innovation.”

This isn’t just about upgrading tools—it’s about rethinking strategy. Fraud models must be agile, continuously learning and capable of adapting to new threat vectors.

Microtrends—subtle, fast-moving patterns—are often missed by older models. Detecting these requires continuous learning loops and real-time behavioral signals. We’ve built systems that can ingest and act on signals like typing speed, geolocation and device usage, enabling personalized risk management and low-friction responses.

Traditional models often rely on historical data and static rules. But fraud is dynamic. Our systems must be able to detect anomalies in real time, adapt to new behaviors and respond with precision, while investing in building digital footprints—not just of users, but of agents acting on their behalf.

Confirmed fraud data is limited, which makes training robust models a challenge. Synthetic data has become a vital tool. It allows us to simulate fraud, accelerate learning and run safe experiments without touching client data. This also opens the door to collaboration—sharing intelligence without sharing raw data.

“Synthetic data has proven to be increasingly vital in building resilient AI models,” Hamilton said. “It’s a key unlock for collaboration—Visa and i2c can share intelligence without sharing raw data.”

We’ve seen firsthand how synthetic data can help us test edge cases, simulate rare fraud scenarios and improve model performance without compromising privacy or compliance.

Fraudsters collaborate. So must we. Sharing intelligence securely and responsibly is essential to building resilience across the ecosystem. We’re investing in infrastructure that supports federated models and tokenized data sharing—so institutions can collaborate without compromising privacy.

“Fraudsters are exploiting ecosystem gaps,” Pruss said. “Sharing intelligence among credit unions, banks, merchants and regulators strengthens defenses.”

Across our industry, it’s essential that we move beyond sharing data—and start sharing knowledge. That means building protocols, infrastructure and trust across institutions, sectors and borders.

Responsible AI governance is non-negotiable. We must ensure transparency, consent and traceability in how data is used. That means building systems that are explainable, auditable and adaptable to user preferences.

We’ve built our fraud systems with these principles in mind—ensuring that users remain in control and that every decision made by AI can be traced, explained and adjusted.

Commerce is inherently local—even when payments are global. Fraud detection must reflect that. We’ve invested in localized feature engineering and segmented model training to ensure relevance and accuracy across geographies, payment rails and user profiles.

There’s no one-size-fits-all model. What looks suspicious in one region may be normal in another. That’s why our models are tuned to local behaviors, transaction types, and user profiles.

If you’re a credit union executive looking to modernize your fraud strategy, start with business objectives. Don’t just upgrade the detection piece—upgrade the entire member experience.

Pruss says it’s imperative to treat fraud strategy as a living system, not a one-time project. Modernizing fraud infrastructure requires continuous monitoring, real-time data processing and cross-functional collaboration.

“Phishing accounts or attacks have now accounted for about 80% of cybersecurity incidents in the financial sector and over 50% of fraud now involves AI,” he said.

And the impact is lasting: a 2025 survey found that nearly 6 in 10 fraud victims were likely to reduce or terminate their banking relationship with the affected institution. 

We’ve seen the best results when fraud strategy is aligned with member experience, operational efficiency, and long-term growth. It’s not just about stopping fraud—it’s about building trust.

AI is not something you install and forget. It requires constant tuning, retraining and governance. Fraudsters don’t sleep—and our models shouldn’t either.

“Especially in high-stakes fraud scenarios, human judgment is essential,” Hamilton said. “AI can augment that—but it needs oversight.”

We’ve built our systems to support human-in-the-loop decisioning, especially when context and nuance matter. That’s how we ensure AI remains a tool for trust—not just efficiency.

At i2c, we’re powering the future of fraud defense with bold innovation and trusted reliability. Our unified, AI-driven platform combines next-gen agility with decades of proven expertise—helping credit unions detect threats faster, personalize protection and scale securely. From microtrend detection to agentic payments, we’re not just keeping pace with fraud—we’re outsmarting it. For more details, visit i2cinc.com or connect with me to learn how our configurable, composable solutions can help you modernize fraud strategy, elevate member experience and grow with confidence.

Static rules and legacy models can’t keep up with today’s fraud landscape. Agentic AI systems—capable of learning, acting and adapting in real time—are critical for detecting fast-moving threats like deepfakes, synthetic identities and account takeover fraud.

Fraud is dynamic. Your defenses must be too. That means continuous monitoring, real-time data processing and collaboration across teams and institutions. Synthetic data and federated intelligence sharing can accelerate learning and improve resilience.

Transparency, consent and explainability must be built into every AI-driven decision. Fraud prevention should enhance—not hinder—the user experience. Localized model tuning and human oversight are key to maintaining trust and long-term relationships.