By Ray Birch
EAST LANSING, Mich.—For years, credit unions have invested heavily in tools to stop card fraud, account takeovers, and suspicious transactions. Increasingly, however, the front line of fraud isn’t the debit card or the ACH file—it’s the phone call.
At Michigan State University FCU, that reality has driven a significant upgrade in how the $8-billion institution authenticates members and evaluates risk during live interactions. Over the past year, the credit union has deployed advanced fraud-detection tools designed to analyze calls in real time, flag high-risk activity, and even identify synthetic—or deepfake—voices before fraud occurs.
The effort reflects a broader shift across the industry, as artificial intelligence lowers the barrier for criminals to convincingly impersonate legitimate members using voice-generation tools trained on publicly available audio.
From Transactional Fraud To Voice Risk
Colleen Cole, vice president of the member service center at MSUFCU, said the credit union already had strong controls around transactional fraud, including card and account activity. What it lacked was clear visibility into attempted fraud over the phone—particularly social-engineering attacks aimed at call-center staff.
To close that gap, MSUFCU partnered with Pindrop Security Solutions to layer voice and call-risk intelligence into its authentication process. The tools evaluate multiple risk signals, including call origin, device characteristics, voice patterns, and behavioral indicators, giving agents real-time insight into whether a call warrants stepped-up—or relaxed—authentication.
The technology doesn’t rely on a single red flag, Cole said. Instead, calls are scored based on numerous factors that together indicate elevated risk, including signs of synthetic voice usage. That intelligence allows agents to slow down interactions, add verification steps, or halt activity entirely before damage is done.
Stopping Fraud Before It Becomes A Loss
Since implementing the tools roughly a year ago, MSUFCU has identified hundreds of high-risk cases involving attempted account takeovers or fraudulent inquiries. By intervening early, the credit union estimates it has protected $2.57 million in member balances that could have been compromised had those calls succeeded.
Importantly, Cole emphasized that the savings weren’t tied to a single dramatic incident. Instead, they reflect an accumulation of stopped attempts—exactly the kind of fraud that often goes undetected until funds are already gone.
Before deploying the tools, MSUFCU had no reliable way to quantify how much fraud was being prevented at the call-center level, Cole said. Now, it can measure not only losses avoided, but also efficiency gains, including reduced call-handling time and better agent confidence—benefits that more than offset the annual licensing cost of the technology, she said.
How Deepfake Fraud Is Evolving
Fraudsters are increasingly using AI to generate realistic voices by scraping audio from social media, voicemail greetings, and other public sources. During live calls, criminals can use AI tools to quickly produce responses that sound like the real member, even when answering security questions.
While MSUFCU has not yet encountered video-based deepfake attacks, Cole said the technology is advancing quickly, and voice fraud alone is becoming more sophisticated and more accessible to criminals.
Industry data underscores the urgency. Firms such as TransUnion and Javelin Strategy & Research have warned that generative AI is accelerating social-engineering fraud, while vendors report double-digit increases in voice-based attacks against financial institutions. Analysts note that call centers—where human trust is still central—are especially vulnerable.
A Message For The Industry: Act Early
Cole’s advice to other credit unions: Don’t wait until deepfake fraud becomes obvious or costly.
The growth rate, she said, is steep enough that institutions not feeling the impact today are likely to encounter it within the next year. Credit unions that fail to invest proactively may find themselves reacting to losses rather than preventing them, she said.
As fraud shifts from stolen cards to stolen identities—and from physical theft to synthetic impersonation—MSUFCU’s experience offers a clear lesson: Advanced detection tools are no longer optional. They are quickly becoming a baseline requirement for protecting members, staff, and trust in an AI-driven threat landscape.