Voice Cloning Scam: What to Do When Your Cloned Voice Calls Your Bank or Family

Last updated: June 2026

A client called us last winter at 2 a.m. His mother had just wired ninety thousand dollars to a “lawyer” because she heard her son’s voice on the phone, sobbing, telling her he had hit a pedestrian and needed bail. The voice was his. The cadence, the nickname he uses only for her, the slight stammer when he is nervous. None of it was him. A thirty second clip of a podcast he had recorded two years earlier had been enough for the synthesizer.

This is the part of AI reputation management that nobody warns you about. Your search results can be clean, your social profiles can be locked down, your Wikipedia page can be accurate, and a stranger can still steal your voice from a YouTube interview and use it to drain your family’s savings, impersonate you to your CFO, or call your bank and ask for a wire transfer in your name.

This post is the playbook we run for clients in that situation. It covers what to do in the first hour after a voice clone surfaces, how to handle bank impersonation, family impersonation, and workplace impersonation, what the new federal and state rules actually let you do, and how to harden your voice so the next clip the criminals find is worth less to them.

It is not legal advice. It is the operational sequence we use when a client tells us their voice is on a call they never made.

How Voice Cloning Actually Works in 2026

Two years ago, a usable voice clone required several minutes of clean audio and access to a research model. Today, a passable clone takes three to fifteen seconds and runs on a consumer laptop. The technology is no longer the bottleneck. The bottleneck for criminals is matching the clone to a believable script and a believable callback path.

That changes the threat shape for every public-facing professional. Anyone who has done a podcast, a webinar, an investor call, a deposition, a video voicemail, or a TikTok is sample-rich. The Federal Trade Commission’s Voice Cloning Challenge, which ran in 2024, was explicitly framed around the fact that cloning detection had to advance because cloning capability already had.

The most common attack patterns we see fall into three buckets. Family impersonation, where a synthetic voice calls a parent, spouse, or child claiming to be in a crisis and asking for money. Bank impersonation, where the synthesized voice calls a financial institution and tries to authorize a transfer, change an address, or unlock an account. Executive impersonation, where the voice of a CEO or CFO calls a controller, treasurer, or assistant and authorizes a payment under time pressure.

Each one has a different first move. The shared first move is the same as in every other reputation crisis. Stop talking, start recording.

The First Hour

The shape of a voice cloning incident on hour two is rarely the shape it had on hour one. Money moves, calls multiply, and the criminals iterate as soon as one path closes. Four things should happen in the first sixty minutes.

Preserve everything you can. If anyone in your circle received the call, ask them to save the call log entry, any voicemail, any text messages from the caller, and any wire confirmations. Most US carriers retain call detail records for a limited window, and a written request to the carrier within that window is the difference between having an evidentiary trail and not. The Reporters Committee for Freedom of the Press keeps an up to date reference on subpoena and preservation requests that maps cleanly onto this situation.

Call the receiving institution before the call window closes. If money has moved, the receiving bank’s fraud line is the single most important phone call in the entire incident. Wires can sometimes be recalled in the first hour through the originating bank’s correspondent network, especially if both banks are domestic and the receiving account has been flagged. The Consumer Financial Protection Bureau’s guidance on wire fraud recovery is realistic about how narrow that window is.

Notify the people you are connected to publicly. A criminal who has cloned your voice for one mark will often try the same clone on others in your circle. A short, plain message to your closest family, your office, and any board or executive assistant inside one hour cuts the next attempt off before it lands. Tell them the call did not come from you. Give them a code word they should use to verify any future request involving money.

Loop in counsel and law enforcement. Voice cloning fraud falls under the FBI’s Internet Crime Complaint Center jurisdiction when money moves across state lines, which it almost always does. The IC3 complaint is also useful evidence when you later approach the platform that hosted the source audio or the model that produced the clone. State attorneys general with elder fraud or financial fraud units are the standard second call. For an executive impersonation case, your general counsel and your insurance carrier need to be on the chain inside the first hour, because most cyber and crime policies have notice windows that start the moment a covered employee learns of the loss.

Bank Impersonation: When a Voice That Sounds Like You Calls Your Bank

In a bank impersonation case, the criminal uses a clone of your voice to call your financial institution’s customer service line and authorize something on your account. The most common asks are a wire to a new beneficiary, an address change, a debit card reissue to a new address, or removal of a hold. Voice biometrics on the receiving side were supposed to defeat this, and they often do, but cloned voices now defeat naive voiceprint systems at meaningful rates, which is why every major bank is migrating to liveness checks and multi factor confirmation as the default.

If you discover that someone has used a clone of your voice on your bank, here is the sequence.

Open a fraud claim with the bank immediately and ask them to flag the account for voice authentication review. Request that any voiceprint enrolled on the account be invalidated and re-enrolled with a liveness check. Ask whether the bank can move the account to a passphrase or token based verification system for the next thirty days. Most major US banks now offer this on request.

File the Federal Trade Commission fraud report at ReportFraud.ftc.gov. Under the Fair Credit Billing Act and Regulation E, consumers have meaningful protections against unauthorized electronic transfers, but the clock for asserting those rights is short and depends on you reporting promptly.

Pull your credit reports and freeze them. The clone may not stop at one institution. The Consumer Financial Protection Bureau’s identity theft guide covers the freeze process.

Ask the bank for a copy of the audio recording of the fraudulent call. Most institutions record customer service calls and will produce the audio to the account holder on request, especially when fraud is alleged. The recording is evidence and a useful artifact for any later civil claim.

If the call originated from a US number, file a complaint with the Federal Communications Commission. The FCC’s February 2024 declaratory ruling on AI generated voice calls made AI generated voice calls illegal under the Telephone Consumer Protection Act when they are unsolicited. That ruling gives state attorneys general a clean cause of action against the originator, and it gives consumers a private right of action with statutory damages.

Family Impersonation: The “Grandparent Scam” With a Real Voice

The grandparent scam predates voice cloning by decades. What is new is that the voice on the line is now actually the grandchild’s voice, lifted from social media, scraped to train a small model, and replayed against a script that names the people the grandchild actually knows.

The Federal Trade Commission’s consumer alert on family emergency voice scams covers the pattern. McAfee’s 2023 research on consumer experiences with AI voice cloning, summarized in the report “Beware the Artificial Imposter”, found that one in four adults had either personally experienced a voice cloning scam or knew someone who had. Two years later the share is higher and the average loss per incident is higher.

If a family member has been targeted, the sequence is the same as bank impersonation with two additions.

The first is the verification code. Every family with public-facing members should have a verbal verification code that any family member can ask for if a call about money, an arrest, an accident, or a medical emergency comes in. The code is not a password. It is a question with an answer only the real person knows, ideally something that does not appear in any social media post or news article. Set the code now, before you need it.

The second is the local police report. Even if the loss is small, a local report establishes the chain of custody for the evidence, gets the case onto the state attorney general’s radar, and unlocks the AARP, state elder fraud unit, and adult protective services referrals that are sometimes more responsive than federal channels. AARP’s fraud watch network maintains a state by state directory of fast intake lines for elder impersonation cases.

For minors targeted in family impersonation, the National Center for Missing and Exploited Children’s CyberTipline takes voice cloning intake when the victim is under eighteen.

Executive Impersonation: When a Voice That Sounds Like the CEO Calls the CFO

The corporate version of this attack has been around since at least 2019, when criminals used a synthesized executive voice to authorize a quarter million dollar transfer at a UK energy firm. Six years later the attack runs faster, on cheaper compute, and against a wider universe of mid market companies that did not exist on the criminals’ radar in 2019.

The control that defeats this attack reliably is procedural, not technological. No transfer above a threshold should ever clear on the basis of a phone call alone, no matter how convincing the voice. The standard control is dual approval through a different channel than the request arrived on. If the CEO calls and asks for a wire, the controller confirms by video on a separately initiated meeting, by a passphrase set up in advance, or by an in person check.

If a clone of an executive’s voice has already been used against a company, here is the response.

Lock the requested transaction and any in flight transactions linked to the impersonated executive. Notify the bank’s fraud team, the company’s external counsel, the cyber insurance carrier, and the audit committee chair. The Securities and Exchange Commission’s cybersecurity disclosure rule, in effect since December 2023, requires public companies to disclose material cybersecurity incidents on Form 8-K within four business days of a materiality determination. A successful voice clone executive impersonation that results in a material loss can trigger that obligation. Your general counsel needs to make the materiality call early so the disclosure clock is not running silently.

Preserve the originating call detail records, any recorded voicemail, and the email or chat thread that accompanied or referenced the call. The Verizon Data Breach Investigations Report catalogs the rise of social engineering and pretexting in financial fraud, and the standard post incident investigation pattern maps cleanly onto a voice cloning case.

Notify your customers and counterparties only after counsel approves the language. Cloned voice attacks usually try multiple targets inside the same organization, so a quiet, fast internal alert to every employee with payment authority is more useful than a public statement.

If the source audio used to train the clone came from a public appearance, podcast, or earnings call, document where it came from. The pattern of source selection is sometimes useful intelligence about who is behind the campaign.

What the Law Actually Lets You Do

Voice cloning sits in an awkward legal seam. The technology is new, the harms are not. Most of what plaintiffs and prosecutors are using right now is existing law applied to a new fact pattern, with a small number of new statutes filling the gaps.

The FCC’s February 2024 declaratory ruling made AI generated voice calls illegal under the TCPA. That gives state attorneys general direct enforcement authority and gives consumers a private right of action with statutory damages of up to fifteen hundred dollars per willful violation.

The TAKE IT DOWN Act, signed in May 2025, created a federal removal obligation for non-consensual intimate imagery including AI generated content. The act covers visual depictions specifically, so voice only cloning is not within its core mandate, but a voice clone paired with synthetic imagery is.

The pending NO FAKES Act, reintroduced in 2024 with bipartisan support, would create a federal right of action against unauthorized digital replicas of a person’s voice or likeness. As of June 2026 it has not yet passed, but eleven states have enacted voice or likeness replica statutes that work in the meantime. Tennessee’s ELVIS Act, California AB 2602 and AB 1836, and right of publicity statutes in New York, Illinois, Texas, and Virginia each give voice cloning victims a viable civil cause of action when the clone is used commercially or in a way that damages the person depicted.

The Federal Trade Commission’s final rule on the use of impersonation, which took effect in 2024 and was strengthened in 2025 to cover impersonation of individuals, gives the FTC direct enforcement authority and the ability to recover money for victims.

The European Union AI Act requires generative voice systems to label outputs and gives EU residents a removal path when the content is hosted or accessed inside the union. The UK Online Safety Act extends similar duties to platforms operating in the UK.

If the original training audio was scraped without your consent, copyright and right of publicity claims may overlap. We covered the broader removal framework for synthetic media in a separate post on how to remove a deepfake of yourself earlier this year. The voice equivalent uses the same underlying levers, with the additional FCC and TCPA hook that does not apply to silent video.

Hardening Your Voice From Future Cloning

Once a voice clone of you is in the wild, it does not get recalled. The only durable response is reducing the value of any future clone and making verified communication the default for the people who would actually transact with you.

Two practical steps matter most.

The first is the verification code we mentioned above, extended to every relationship where money can move on your say so. Your spouse, your parents, your adult children, your assistant, your CFO, your controller, your wealth manager, and your bank’s private client desk should each have a passphrase or callback procedure tied to your name. The passphrase has to be a fact that is not findable through public search. The callback has to use a number that was confirmed in person or through a previously established channel, not the number the suspicious call came from.

The second is voice biometrics enrollment at every institution that offers it, paired with the explicit option to require a liveness check. Liveness checks ask the speaker to respond to a randomized prompt that a cloned voice cannot pre-render. Banks, brokerages, and large healthcare systems increasingly offer this; we recommend it.

Beyond those two, watermarking and provenance tools are gaining traction on the supplier side. The Content Authenticity Initiative and the Coalition for Content Provenance and Authenticity (C2PA) publish open standards for tagging synthetic media at the point of creation. The major voice synthesis vendors have signed on to varying degrees. None of these tools defeat a committed criminal who simply strips the metadata, but they raise the friction enough that opportunistic attackers move on.

Finally, audit your public audio. Earnings calls, podcasts, conference panels, video voicemails, and longform interviews are the primary training material for voice clones. You cannot pull them all down, but you can decide that future ones get hosted in places that watermark the output and that your daily voice notes and family voicemails are not posted publicly. Reputational hygiene now includes audio hygiene.

When a Voice Clone Has Already Harmed Your Reputation

Sometimes the call gets through. The money moves, the family panics, the company files an 8-K, and the press picks it up. At that point, the problem is no longer just operational. The story now lives on Page 1 of search results and inside the AI chatbots that summarize you for anyone who asks.

This is the seam where voice cloning crosses into the rest of our work. The press coverage of a successful clone is often more harmful than the clone itself. A single article that says a public figure was “fooled by an AI voice scam” can rank above their own website for their name, surface in AI Overviews and Perplexity citations, and follow them into every diligence file for years.

The response sequence is the same one we run for any crisis. Stabilize the story, push accurate first party content into the index, request corrections and clarifications where the coverage was wrong, and use the content removal and search suppression playbook on the remaining URLs that will not move. For executive cases, the executive and individual crisis workflow is built for exactly this fact pattern. For corporate cases, the company crisis management workflow handles the disclosure, vendor, and customer narrative in parallel.

We guarantee outcomes on the removal and suppression side of this work because the methodology has matured to the point that we can. If a piece of negative content is eligible for removal under platform policy or applicable law, we remove it. If it is not eligible for removal, we suppress it below the fold of the relevant search results and AI answers within the timelines we set in writing. That guarantee is the part of our individual reputation management and business reputation management work that we get asked about most.

A Last Word for Anyone Who Records in Public

If you make a living talking, your voice is in the index whether you put it there or not. Podcasts, earnings calls, webinars, conference videos, and longform interviews are the largest single source of voice training material for commodity clone tools. None of that means you should stop talking in public. It does mean that the verification protocols around your name, your family, and your authority to move money have to assume your voice is not a credential.

The clients who recover fastest from voice cloning incidents are the ones who had a verification code in place before the clone arrived, who had a single point of contact at their bank and an assistant trained to insist on dual confirmation, and who had a search results page so cleanly first party that a news story about the incident did not relocate to the top of their results. The clients who recover slowest are the ones who learned all three lessons in the same week.

If your voice is already in a clone you did not authorize, or if you want to put the verification protocols in place before you ever need them, we can help. A short free consultation is the right first step. We will tell you on that call what we can guarantee, what timeline we work to, and what to do in the meantime.