Callback verification is a wire-approval control that confirms a payment request by phoning the requester before funds move. It works only if the voice on the line counts as proof of identity, and voice cloning weakens that assumption. It holds up better when the callback goes to a number already on file, two different people approve, a hold period applies, and each approval ties to a verified person.
- The FBI’s Internet Crime Complaint Center received 24,768 business email compromise complaints, with about $3.05 billion in reported losses, in 2025 (FBI IC3, 2025 Internet Crime Report).
- Callback verification fails when the number comes from the request, when a recognized voice is the only proof, or when the number on file was itself altered.
- Calling a number already on file, requiring two approvers, and adding hold periods hold up better than code words or a familiar voice.
- Presence-aware sign-in confirms who is at the approving workstation; it does not detect a cloned voice or judge whether a request is genuine.
How big is the wire fraud problem, and where does voice cloning fit?
Business email compromise remains one of the largest fraud categories in the United States, and voice cloning is a growing but still small part of it. The numbers below come from the FBI and the credit union regulator, not from a vendor.
In 2025 the FBI’s Internet Crime Complaint Center received 24,768 business email compromise complaints, with reported losses of $3,046,598,558 (FBI IC3, 2025 Internet Crime Report).
The same report notes that voice cloning can be used to request a wire payment, and that businesses reported losses of more than $30 million to business email compromise scams involving AI in 2025. That is a small share of the total. The reason to act is not the size of that figure today. It is that the control most institutions rely on, a phone call to a known person, was designed for a world where a convincing voice was hard to fake.
Credit unions see the pattern directly. The NCUA reports that credit unions filed 588 cyber incident reports between May 1, 2025 and April 30, 2026, a group that includes phishing and email compromises, and that credit unions reported forty known instances of business email compromise in that period, compared with 27 in the previous one (NCUA, 2026 Cybersecurity and Credit Union System Resilience Report).
FinCEN raised the same theme in November 2024. Its alert FIN-2024-Alert004 describes criminals using generative AI to defeat identity verification, and notes that deepfake voices and video are used in business impersonation schemes. Its listed best practices include multifactor authentication, including phishing-resistant MFA, and live verification checks.
Where does callback verification assume a voice is proof?
Callback verification fails at four specific points, and each one treats the sound of a voice as evidence of who is speaking. Walk through your own procedure and check for these.
- The number comes from the request. If the email or the caller supplies the callback number, the attacker answers it.
- The call is inbound and urgent. A finance employee receives a call that sounds like the CEO or a long-standing member. Recognition becomes the check, and recognition is exactly what cloning targets. Public earnings calls, webinars, and conference videos supply plenty of source audio.
- The number on file was changed. If an earlier request altered the vendor or member contact record, calling “the number on file” calls the attacker.
- The voice is the verification. Some procedures ask the person on the callback to confirm a detail by voice alone, or use a voiceprint. Both depend on the voice being hard to fake.
None of this makes callbacks useless. A callback to a verified number on file still adds friction an attacker must overcome. The point is that it should be one layer, not the deciding one.
What changes make wire approval hold up?
Take the voice out of the decision and add checks that do not depend on recognizing anyone. Five changes cover most of the exposure.
- Call back only to a number already on file, and protect that record. Changes to contact details should follow their own verified process with a waiting period before they take effect.
- Require dual approval from two different people. Each approver signs in on their own workstation with their own credentials, so one persuaded employee cannot release funds alone.
- Add hold periods. Apply a delay to new payees, changed payment instructions, and amounts above a set threshold. Time is the cheapest control against urgency.
- Treat code words as a supplement. A fixed, shared phrase can be coaxed out of an employee during a convincing call. It is better than nothing and weaker than the items above.
- Add step-up authentication at the moment of release. Ask for a stronger check tied to a verified, present person, not a one-time code that can be read aloud over the phone.
Finally, train staff on the script itself. The common thread in these attacks is a plausible reason to skip the process just this once. A written rule that no one may waive the process, including executives, removes the burden from the employee on the phone. Our overview of modern cybersecurity technology explains how these checks fit with identity controls.
Does New York’s MFA rule cover this?
It covers access to systems, not the decision to trust a request, so it does not close this gap by itself. Section 500.12(a) of the New York Department of Financial Services cybersecurity regulation, 23 NYCRR Part 500, states that “Multi-factor authentication shall be utilized for any individual accessing any information systems of a covered entity.” Section 500.1 defines MFA as verification of at least two of three factor types: knowledge, possession, and inherence. Under the regulation’s transition schedule, this requirement took full effect on November 1, 2025, and entities that qualify for a limited exemption under 500.19(a) face a narrower scope.
Section 500.12(b) also allows the CISO to approve in writing “reasonably equivalent or more secure compensating controls,” which must be reviewed at least annually. Part 500 applies to entities licensed by DFS. Many credit unions answer to the NCUA instead, so confirm which applies to you.
There is a practical lesson even outside New York. An employee who has signed in with strong MFA and is then persuaded by a cloned voice still passes every access check. Access controls and approval controls solve different problems, and a wire process needs both.
Where does presence-aware sign-in help with wire approval?
It helps with the approver, not the caller. When someone releases a wire from a Windows workstation, sign-in with Proximia requires a live biometric and the user’s own registered XiFi Card, so the person at the keyboard is the authorized person. The session stays open while the card is nearby. When it moves out of range, the workstation locks within seconds, at a distance and timing your institution sets, and signing back in requires the biometric again.
That supports dual approval in a practical way. Each approver has an individual sign-in, so the record shows two distinct people rather than one shared teller sign-in. It applies equally to a shared branch station and to a single-user desk.
Being direct about limits matters here. Proximia Connect does not detect cloned voices or deepfakes, does not judge whether a payment request is genuine, and does not stop an authorized employee who has been convinced to approve. It also does not extend into a cloud-hosted wire platform’s own session after a federated handoff. It narrows one question: whether the approver at the workstation is verifiably the authorized person who is present.
How Proximia Makes Identity Persistent
Proximia replaces point-in-time authentication with Persistent Identity Assuranceâ„¢, a new category of identity and access management. We continuously prove the right person is present.
- Confirm. A live biometric verifies it is you at sign-in, not just someone with a code. Images are never stored or transmitted.
- Bind. Your identity is bound through Proximia’s XiFi® technology. User and device become one.
- Persist. Your presence enables continuous access. Walk up work, walk away lock.
To see how this works at an approving workstation in your institution, request a demo at proximia.com/contact-us.
Frequently Asked Questions
What is callback verification in wire transfers?
It is a control where staff phone the requester, using a trusted number, to confirm a payment instruction before releasing funds. It reduces fraud from email requests, but it assumes the person answering can be identified by voice.
Is a code word enough to stop voice-cloning fraud?
No. A code word is a fixed secret that can be extracted through pressure or pretext during a call. Use it as a supplement to dual approval, hold periods, and calling a number already on file.
Does MFA stop a cloned voice?
No. MFA protects access to systems. If an authorized employee signs in correctly and is then persuaded to approve a fraudulent wire, MFA has done its job and the fraud still succeeds. Approval controls such as dual sign-off and hold periods address that.
What should a small credit union change first?
Start with the contact-detail change process and hold periods for new payees. Both are procedural, cost little, and address the points where attackers most often redirect funds.
Sources
- FBI Internet Crime Complaint Center, 2025 Internet Crime Report, FBI, 2026. https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf (read directly: 24,768 BEC complaints; $3,046,598,558 in reported losses; voice cloning note; more than $30 million in AI-involved BEC losses).
- FinCEN, Alert on Fraud Schemes Involving Deepfake Media Targeting Financial Institutions (FIN-2024-Alert004), U.S. Department of the Treasury, November 13, 2024. https://fincen.gov/system/files/shared/FinCEN-Alert-DeepFakes-Alert508FINAL.pdf
- New York State Department of Financial Services, 23 NYCRR Part 500 (Second Amendment text), sections 500.1, 500.12 and 500.22, DFS, November 1, 2023. https://www.dfs.ny.gov/system/files/documents/2023/10/rf_fs_2amend23NYCRR500_text_20231101.pdf
- National Credit Union Administration, 2026 Cybersecurity and Credit Union System Resilience Report, NCUA, June 2026. https://ncua.gov/news/publication-search/cybersecurity/2026-cybersecurity-and-credit-union-system-resilience-report (588 cyber incidents, May 1, 2025 to April 30, 2026; 40 known business email compromise instances, up from 27). The earlier 2025 report (539 incidents, May 1, 2024 to April 30, 2025) at https://ncua.gov/news/publication-search/cybersecurity/2025-cybersecurity-and-credit-union-system-resilience-report was confirmed and superseded.




