A Familiar Voice Is No Longer Proof Your Member Is Who They Say They Are
Credit unions have always run on relationships, and the phone has always been part of that. A staff member recognizes a voice, trusts it, and helps. The trouble is that a convincing voice can now be produced from a short audio clip, and fraudsters know it.
America's Credit Unions described one institution that began screening incoming calls with AI in 2024. Over a single year, the tool flagged about $2.57 million in fraud exposure from deepfake calls, close to a quarter of a million dollars a month that human ears might have missed. That's one credit union's figure and not an industry average, but it shows the scale of a problem that used to feel theoretical. Deloitte's Center for Financial Services has projected that generative AI fraud losses in the United States could reach $40 billion by 2027.
It isn't only voice
On a recent industry podcast, a SentiLink product leader named romance scams as the biggest threat credit unions face, with fraudsters using AI to scale up, build fake social media profiles, and clone voices to manipulate victims, especially older members. He also described AI helping criminals with check fraud, including stealing checks and using synthetic identities to cash them. America's Credit Unions' own list of 2026 fraud threats includes deepfakes, synthetic identities, authorized push payment fraud, vendor email compromise, and romance scams.
The pattern across all of these is the same. Fraud has become cheaper to attempt at scale, and the old signals, like bad grammar or an odd voice, are disappearing.
What helps in practice
Stop treating recognition as authentication. A staff member who says "that sounds like Mr. Jones" has not verified anything. Make it normal, and expected, to verify through a second channel for anything involving money movement or account changes.
Use callbacks to numbers on file. For higher-risk requests, hang up and call the member back at the number already in your system, not one the caller provides.
Train staff to ask questions. The SentiLink guest recommended front-office and back-office teams work together, with frontline staff trained to notice and question unusual activity. A calm, scripted question like "I'm going to confirm this a different way" protects both the member and the employee.
Add AI where it genuinely helps. Call screening and behavioral analysis can catch patterns people can't hear. They work best as one layer, not the only one.
Talk to members. Older members in particular benefit from hearing, in plain language, that your credit union will never ask for a code over the phone and that a call from a "grandchild in trouble" deserves a pause and a callback.
The balance to strike
The goal isn't to make every call feel like an interrogation. It's to decide in advance which requests need extra checks, so the extra step is routine and nobody has to improvise under pressure. Start by listing your five riskiest phone requests and writing down what verification each one requires.
Sources