Deepfakes Are Fueling Account Takeovers: How Merchants Fight Back

Deepfake account takeover fraud is rising. See what Microblink's new data shows and the layered defenses online merchants can add in 30 days.

Adrian ReyesSenior Payments & Fintech Editor

October 27, 2026 · 8 min read

Person in a black hoodie hides their face behind an ace of spades playing card in a blurred green forest

Photo: Pixabay

On this page 7 sections

What Microblink's Account Takeover Data Shows
How Deepfake Account Takeover Fraud Gets Past Verification
Why Taken-Over Accounts Become Chargebacks and Monitoring Risk
Warning Signs: Password Resets, New Devices and Payment Method Swaps
Layered Defenses: Passkeys, Step-Up Authentication and Liveness Checks
Telling a Customer's AI Agent From an Attacker
A 30-Day Account Security Checklist for Online Merchants

Key takeaways

Microblink's 30% figure counts fake documents among U.S. ID fraud failures. Altered real IDs (photo forgery, 60%) were the larger share.
Stolen passwords still drive most attacks. Deepfakes target account recovery, payout changes and withdrawals, so harden those flows first.
One takeover incident on stored cards can produce enough fraud disputes to move a month's chargeback ratio.
Let verified AI agents buy, but require the human for any change to credentials, contact details, payouts or payment methods.

Deepfake account takeover fraud works because most businesses check identity once and trust it from then on. Attackers now use AI-generated or face-swapped IDs and synthetic selfie video to pass the checks that guard account recovery, payouts and withdrawals. Then they spend the saved card, wallet balance or payout account as if they owned it. The fix is to verify again at the moments that move money: give customers passkeys, step up authentication before sensitive changes, use liveness checks that catch injected video, and decide in advance what an AI agent may do inside a customer's account.

What Microblink's Account Takeover Data Shows

Digital Transactions reported on Oct. 26, 2026, that Microblink's latest review of account-takeover data found 30% of fraud failures involved fake U.S. identity documents. It also reported that AI and deepfakes are making it easier to impersonate real customers. Microblink is a fraud-prevention firm based in Brooklyn, N.Y.

The wording matters. In Microblink's own statistics post, the 30% figure covers fraud failures involving U.S. identity documents that were flagged as fake documents. A bigger share, 60%, was photo forgery: a real document with the photo manipulated. So the most common bad ID in Microblink's U.S. data was a genuine document altered to fit the attacker, which is harder to catch than an obvious fake. These numbers come from one vendor's own verification traffic. They are not a national rate.

The same post pulls in wider numbers. It cites Javelin Strategy & Research estimates that more than 6 million U.S. consumers had an account taken over in 2025, up 18% from the year before, with losses above $15 billion. It also cites Microsoft findings that more than 97% of the identity attacks Microsoft observed were password attacks. In password-spray attacks, 85% of the targeted usernames had already appeared in known credential leaks.

Read those numbers together. Stolen passwords still open most accounts. Deepfakes and fake IDs matter at the backup doors: the recovery flow, the identity check before a withdrawal, the support agent who can reset an email address. Microblink's December 2025 Fraud Lab findings, reported by Crowdfund Insider, showed a 23% rise in AI-generated fake IDs over 12 months. Face-swap document fraud jumped 200% year over year. CEO Hartley Thompson said anyone online can now produce "a convincing deepfake document using a publicly available GenAI tool." For a broader look at preparing checkout, staff and dispute controls, see our guide to merchant defenses for a 'scary year' of AI fraud.

How Deepfake Account Takeover Fraud Gets Past Verification

Microblink lists the techniques it sees: face-swapped IDs built on stolen real identities, AI-generated photos blended onto real ID templates, fabricated documents whose security features look correct, and injection attacks. In an injection attack, the fraudster feeds fake images or video straight into the capture step instead of pointing a real camera at a real face. A liveness check built only to confirm movement on screen may never notice that the video didn't come from a camera. This is fake ID document fraud built for remote checks, and it scales because the tools are cheap.

Attackers aim these tools at the moments when a business assumes the person is who they say. Account recovery is the obvious one ("I lost my phone, here's my license and a selfie"). Others include the identity check a gaming or sweepstakes platform runs before paying out winnings, the payout-account change a marketplace seller requests, and the chat agent who updates an email because the caller knows the last four digits of the card.

The Microblink fraud report puts the problem in one line. Account creation, identity verification and multifactor setup are all good events, but "none of those events guarantees that every future interaction with the account is legitimate." A customer who passed verification three years ago tells you nothing about who is logged in tonight.

Why Taken-Over Accounts Become Chargebacks and Monitoring Risk

When an attacker spends a victim's saved card, the real cardholder sees charges they didn't make and disputes them as fraud. You lose the goods, the shipping and the dispute fee. The order history also works against you: a sale placed from a new device and shipped to a new address is weak evidence that the customer authorized it. Fraud disputes and issuer fraud reports also feed card network monitoring. Our explainer on Visa's VAMP fraud and dispute thresholds covers how that program counts them.

Here is an illustrative example. A subscription box store processes 2,000 card transactions a month. One weekend, an attacker takes over 25 accounts with stored cards and places one $90 add-on order in each. Assume product cost of $36 per order, $8 shipping and a $25 dispute fee (fees vary by processor). If all 25 are disputed, the store refunds $2,250 through chargebacks. On top of the lost sales, it is out $1,725 in product, shipping and fees: 25 × ($36 + $8 + $25). Those 25 disputes also equal 1.25% of the month's transactions from a single incident. Compare that with the limits in our guide to calculating your chargeback ratio.

The damage looks different by business model:

Subscriptions and memberships: attackers upgrade plans, buy gift subscriptions sent to their own email, or change the login email so the owner can't get back in to cancel.
Online marketplaces: a hijacked seller account sends payouts to a new bank account. Hijacked buyer accounts spend stored cards with sellers who never see the warning signs.
Online gaming, sweepstakes and sports betting: wallet balances and winnings get withdrawn to a new payout method, and the identity check before withdrawal is exactly where a fake ID shows up. Whether these products are legal varies by state, and the operator's compliance duties go well beyond card rules.

Warning Signs: Password Resets, New Devices and Payment Method Swaps

No single signal proves a takeover. What matters is combinations, especially inside one session or one day:

A password reset followed within minutes by an email or phone number change
A new device or location plus a new shipping address plus a larger-than-usual order
A new payment method added to an older account and used at once, or several cards tried in a row
A payout or withdrawal destination changed shortly before a cash-out
Repeated failed identity checks, or the same face appearing on IDs with different names across accounts
Support requests pushing for an urgent email change while the account owner's usual channel stays silent

Microblink's guidance on behavior signals adds rapid or inconsistent data entry, unusual session lengths, and biometric data that doesn't match the device data. Location is another strong layer, covered in our primer on using location data to stop scams.

Layered Defenses: Passkeys, Step-Up Authentication and Liveness Checks

Passkeys replace the password with a cryptographic key stored on the customer's device and opened with a fingerprint, face or PIN. There is no reusable secret to phish or stuff, which addresses the password attacks Microsoft says dominate. Offer passkeys as an option at login and checkout. Forcing them on everyone at once will lock out some real customers.

Step-up authentication means asking for more proof only when the action is risky. Good triggers for account takeover prevention in ecommerce include changing an email, phone number, password or payout account. Adding a payment method, shipping a large order to a new address and withdrawing a wallet balance should trigger it too. For card payments, run 3-D Secure authentication on risky online transactions, such as the first use of a newly added card. Depending on network rules and the authentication result, it can also shift liability for some fraud chargebacks to the card issuer. Confirm the details with your processor.

Liveness and document checks belong in recovery and withdrawal flows, not only at sign-up. When you evaluate a vendor, ask whether it detects injection attacks and AI-altered documents. Also ask whether it has taken part in independent testing such as the U.S. Department of Homeland Security's Remote Identity Validation Rally.

Three cheaper controls close the gaps the tools leave:

Send every credential change notice to the old email and phone, with a one-click "this wasn't me" link.
Hold payout and withdrawal changes for a cooling-off period before money can move.
Never let an ID upload alone reset an account that is protected by a passkey.

Our look at what AI fraud detection catches and misses explains where scoring models fit alongside these rules.

Telling a Customer's AI Agent From an Attacker

Some of the "bots" logging into customer accounts will soon be real shopping agents. Visa introduced its Trusted Agent Protocol on Oct. 14, 2025, developed with Cloudflare. According to Visa's GitHub documentation, it gives an AI agent a cryptographic way to prove who it is and that it's authorized: it presents a digital signature with each interaction. The protocol answers three things for the merchant: whether the agent is recognized and trusted, whether it is acting for a specific signed-in user, and whether it carries valid instructions for this purchase. Akamai has announced support for the protocol in its bot and abuse protection.

AI identity fraud puts merchants in a bind here, because blocking all automation also blocks good customers. Microblink describes the job as sorting each session into a human, an AI agent acting for a customer, or a malicious bot. A practical policy for merchants:

Treat a user-agent string that says "AI agent" as a claim. Proof comes from the signature, checked by your CDN, bot-management vendor or gateway.
Let verified agents browse, build carts and check out with tokenized card details.
Require the human, through step-up, for anything that changes the account itself: credentials, contact details, payout destinations or new payment methods.
Treat unsigned automation hitting your login page at volume as credential stuffing, whatever it calls itself.

That split lets a real customer's agent buy while the keys to the account stay with the customer. For the network plans behind it, see our coverage of Visa and Mastercard's AI shopping agent programs.

A 30-Day Account Security Checklist for Online Merchants

Most of this plan is process work rather than new software. It tightens customer account security without adding friction to every login.

Days 1–7: Write down every way an account can be recovered or changed, including support scripts. Turn on change notices to old contact details.
Days 8–14: Add step-up triggers for credential, payout and payment-method changes. Add a cooling-off hold on payout changes.
Days 15–21: Review fraud rules with your processor and payment gateway: velocity limits on new cards, 3-D Secure on high-risk orders, address-change rules. Start offering passkeys.
Days 22–30: Test your recovery flow the way an attacker would, using a new device and a staff member's ID with their permission. Set your AI agent policy and a takeover response plan: freeze, notify, refund and reset.

If you do only one thing this month, run the recovery test in step four, because it shows which door an attacker would use first. If your setup for online payments or high-risk processing needs a review, MannyPay specialists respond to free quote requests within one business day.

Sources

About the author

Adrian Reyes

Adrian Reyes is Senior Payments & Fintech Editor for Beyond the Swipe, covering credit card processing, merchant accounts, payment technology and processing fees.

This article is for general educational purposes and is not legal, tax or financial advice. Card network rules and state laws change; confirm current requirements for your business.