Deepfake Fraud: How It Bypasses Merchant Verification
Deepfake fraud puts a fraudster through your identity check as somebody else, using AI-generated video, voice, or images. The damage starts well before a card is ever charged.
I've worked chargeback disputes where the account itself was the fraud, not the card, and deepfakes now account for one in five fraud attempts against biometric identity checks.
By the end of this you'll know what to change at your signup flow and your support desk.
Key takeaways
Worried the account behind a charge was never real? Our dispute alerts reach you while a refund still closes the dispute.
What is deepfake fraud?
Deepfake fraud is the use of AI-generated video, voice, or images to impersonate a real person and pass an identity check. The same tools also talk a person into acting for the fraudster.
What makes it work now is quality and speed. A model can build a face or a voice clip that survives a manual look or a live phone call.
Every check that assumes someone will notice starts failing there.
One 2019 case shows the shape of it.
A UK energy firm's chief executive wired €220,000, roughly $243,000, to a supplier, reported by Forbes. He'd taken a call in a cloned version of his boss's voice.
Deepfake fraud is a technique, and it gets pointed at fraud that already existed. Synthetic identity, account takeover, and social engineering all predate it, which our generative AI fraud guide covers in full.
Where deepfake fraud actually hits a merchant
Deepfake fraud reaches a merchant at your signup identity check and at your support desk. At both, one signal decides whether someone is who they claim:
- Signup verification, where a synthetic face or ID clears onboarding.
- Support contact, where a cloned voice gets account access changed.
Signup is the one most merchants can fix this week.
1. Bypassing identity verification at account signup
A deepfake at signup gets a fraudster a real account on your platform, opened in someone else's name. Your onboarding asks for a selfie, a short video, or an ID photo. An AI-generated face answers all three.
The fraud happened before the first order, so the account looks ordinary in your admin panel:
- A verified badge from the check it already passed.
- A valid payment method on file.
- A clean order history with no disputes.
Account takeover hijacks a real customer's account. A deepfake signup creates one whose named person was never your customer.
Your fraud scoring won't help much either. Most order-level scoring asks whether the account is new and whether the card matches the buyer.
This account already passed your own identity check.
2. Social engineering a customer-support team
A cloned voice on a support call gets your agent to hand over the account, because your agent is the verification step. In our support conversations, the recurring theme is a caller who knows the account holder's name and last order.
Support teams are built to resolve things fast, and that pressure is the weak spot. An agent measured on handle time has every reason to believe a frustrated caller who sounds like the customer.
Once the agent approves a reset, the next orders ship to a new address on a saved card. The real cardholder reads a statement weeks later and files them as fraud.
Video verification moved the same problem to a screen.
A live face proves less than it did two years ago, so treat a video check as one signal.
How common is deepfake fraud right now
Deepfakes account for one in five fraud attempts against biometric identity checks. That figure comes from Entrust's 2026 Identity Fraud Report, reported by Biometric Update. It draws on more than a billion identity checks.
Two more figures from that report show what grew fastest. Deepfaked selfies rose 58% in 2025, and injection attacks rose 40% year over year.
Both numbers rose for the same reason, which is cost. Building a convincing face got cheap enough for consumer scams, and the same cheap output beats an automated selfie check.
An injection attack feeds a doctored image or video straight into the system, bypassing the camera altogether.
Read those figures for what they measure. They count attempts against identity systems in banking, fintech, and marketplaces.
Your own risk depends on how much of your signup rests on a photo step.
What actually stops deepfake fraud at signup and support
Pair every identity check with a signal the fraudster doesn't control, device history at signup and a callback at support. The 2-Point Deepfake Exposure Check asks one question about each:
- Signup verification: Does it pass on a single photo or video check alone?
- Support verification: Can an agent change account access from a call alone?
Answer yes to either one and that check is exposed. A signal the fraudster can generate is being treated as proof.
1. Signup verification
Pair every photo or video check at signup with a device and session signal you recorded yourself. Your verification vendor only sees the image the fraudster supplied. Your own logs see the browser, the IP, and how the session behaved.
Three things to check on your signup flow this week:
- Whether the same device or IP opened other accounts recently.
- Whether the session reported a virtual camera or remote-desktop client.
- Whether the ID photo and the selfie arrived seconds apart.
A selfie check passing on a device that opened four accounts this month is worth a manual review.
The limit here is real. Device signals catch volume attacks well, and a patient fraudster on a clean device still gets through.
Send any signup that trips one of those three checks to manual review instead of declining it.
2. Support-desk verification
Call the customer back on the number already on file before you change any account access. A fraudster controls what you hear on an inbound call. The phone, the inbox, and the logged-in session stay with the real customer.
Give your agents three rules they can follow under pressure:
- Send account-change confirmations to the address on file, always.
- Call back on the stored number for any access change.
- Have the customer approve the change from inside their account.
The callback works because the real customer is holding the phone you dial.
Write one named role into the script too, like the fraud lead on shift. Tell agents to transfer any caller who refuses a callback, so refusing never depends on winning an argument.
What catches deepfake fraud after verification fails
A pre-dispute alert catches the charge that follows a deepfake bypass, while a refund still closes it. No control holds against every targeted attempt, so a deepfake-verified account eventually places an order the real cardholder disputes.
Now you're facing true fraud, with a verification record saying the buyer checked out.
The dispute still arrives under the ordinary fraud code, so confirm which one landed with our reason code lookup tool.
A chargeback alert reaches you after the cardholder disputes the charge, before it converts to a chargeback. Refund inside that window and no chargeback gets recorded, so your loss stops at the sale.
Those notices come from two networks:
We connect to both networks and match each alert to the transaction in your processor. The refund happens without anyone watching a queue.
Some deepfake-verified accounts never produce a disputed charge. One abusing a promo or reselling access is a trust-and-safety problem alerts won't see.
How we sourced our data
The support pattern in this article comes from a 12-month review of anonymized, de-identified Chargeback.io support conversations. We report recurring themes across those tickets, counted by conversation volume.
The themes we report describe verification and setup confusion generally, because the ticket data carries no deepfake tag. Reading them as a deepfake exposure is our own analysis.
FAQ
Does deepfake fraud get its own reason code?
No. Visa and Mastercard publish no code for AI-generated fraud, so the dispute arrives under the ordinary fraud code.
Is deepfake fraud only a large-platform risk?
No. Automated attacks run against whatever signup flow is cheapest to beat, and a single selfie check is often the softer target.
Deepfake fraud vs. ordinary identity theft
Identity theft uses a real person's stolen details, while deepfake fraud builds the face or voice that makes those details pass. One supplies the data, the other supplies the proof.
Do alerts cover a deepfake-verified account?
Alerts cover any dispute a participating issuer files, fraud included. Coverage depends on the issuing bank taking part in the network.
