Deepfake KYC Bypass: How Attackers Beat Liveness Checks

By Owen Thurm Published November 6, 2026 6 min read

Quick answer. Yes, against weak checks. FinCEN says criminals have opened accounts with suspected AI-made identities, and a kit reported in April 2026 targets crypto platforms' KYC. Defend in layers: detect injected video, check documents, watch behavior after onboarding.

Watch: Deepfake KYC Bypass: How Attackers Beat Liveness Checks 1:41

Video transcript

Can deepfakes beat KYC at a crypto exchange? Against weak checks, yes. FinCEN says criminals have opened accounts with suspected AI-made identities.

Remote KYC usually checks a document, a live face, and a match between them. Attackers must beat all three, and AI has made faking cheap.

A deepfake has two ways in: presentation attacks show it to a real camera, and injection attacks feed it through a virtual camera or emulator.

ENISA called video injection the most promising attack method, according to market players. NIST says liveness checks make it harder, but cannot stop every case.

In April 2026, Cointelegraph reported a face-swap kit allegedly built to bypass KYC at banks and crypto platforms. Reports cite no platform confirming a bypass.

No single signal proves a deepfake. Watch for virtual cameras and webcam plugins, documents that do not match, odd device data, and withdrawals right after deposits.

Defend in layers. Prove the camera is real, randomize liveness, check the document itself, step up on risk, and re-review account files later.

Ask your KYC vendor which injection routes it detects. If the client app can report verified on its own, no liveness model protects you.

Guardian's fintech pentests attack the flows that authenticate users, authorize actions and approve withdrawals. Read the full article, or talk to Guardian at guardian.security.

Remote onboarding asks a phone camera to prove who someone is. Generative AI has made the face, the document and the video cheap to fake.

In November 2024, FinCEN issued an alert on deepfake fraud schemes (FIN-2024-Alert004). Suspicious activity reports describing deepfakes had risen since 2023.

In June 2026, Hong Kong's SFC warned licensed firms that AI tools may "significantly lower the technical barrier" for deepfake impersonation and phishing. Our SFC circular explainer covers what it expects.

How Bypass Kits Work

Remote identity proofing usually checks three things: an identity document, a live face, and a match between the two.

An attacker has to beat all three. FinCEN describes the raw material it sees:

ID images altered or generated with AI, such as driver's licenses and passports.
Synthetic identities that pair AI images with stolen or fake personal data.
AI-made photos and videos used to get past identity verification.

The live face is the hard part. ENISA, the EU cybersecurity agency, noted in a 2022 report that deepfake videos and "interactive deepfake puppets" can now be created in real time.

In that mode, the attacker's own camera feed is swapped for the target's face as they move. ENISA warns that random prompts "could be ineffective" here, because a real person is responding.

Then the fake has to reach the check. ENISA says a deepfake has only two ways in: shown to the camera, or injected into the camera feed.

Injection vs Presentation Attacks

	Presentation attack	Injection attack
Route to the check	Shown to a real camera	Inserted between camera and verifier
Examples	Printed photo, screen replay, 3D mask	Virtual camera, emulator, webcam plugin
Main defense	Presentation attack detection (liveness)	Genuine-camera checks and injection detection

NIST describes injection attacks as inserting "modified or forged media between the capture point" and the system doing the comparison.

ENISA's report called video injection "the most promising attack method, according to market players". It is not a presentation attack at all: it goes around the camera.

That is why liveness alone falls short. ENISA says passive liveness "may be not as effective" against injected photos, video reprojection and deepfakes.

NIST agrees that liveness and live document capture make injection harder. But it says they "are not sufficient to address all possible cases of these kinds of attacks."

Known Kits and Tactics

Webcam plugins and glitches

FinCEN says some verification tools flag third-party webcam plugins, "which can let a customer display previously generated video rather than live video".

It also describes a social tactic: claiming repeated technical glitches, or asking to switch communication method mid-check, to avoid live verification.

JINKUSU CAM

In April 2026, Cointelegraph reported a fraud kit allegedly sold by a threat actor known as Jinkusu.

It said the kit is "designed to bypass Know Your Customer (KYC) checks at banks and crypto platforms".

Security firm Vecert Analyzer said it uses InsightFace for real-time face swaps, with voice modulation.

Biometric Update called it "a live deepfake tool that is designed to defeat remote identity verification".

These capabilities come from cybercrime trackers, not from tests. Neither report cites a platform confirming a bypass.

Hiring fraud

Face swaps reach beyond onboarding. The FBI said in January 2025 that North Korean IT workers have used AI and face-swapping technology in video job interviews.

See how to spot North Korean IT workers for the hiring checks.

Detection Signals

No single signal proves a deepfake. FinCEN asks institutions to weigh the surrounding facts before deciding activity is suspicious.

At capture

A virtual camera, device emulator or jailbroken device (NIST's examples).
A third-party webcam plugin during a live check (FinCEN).
Signs of manipulation, or signatures of known deepfake tools (NIST).
An image that breaks on fast movement or a hand across the face (ENISA).

In the documents

A photo with visual tells of alteration, or at odds with the stated age.
Several identity documents that do not match each other.
An ID photo that matches a gallery of AI-generated faces.

After onboarding

Device or location data inconsistent with the identity documents.
Logins from IP addresses that do not fit the customer's profile.
Coordinated activity across several similar accounts.
Rapid transactions from a new account, or withdrawals right after deposits.
A customer who declines multifactor authentication.

For a crypto platform, withdrawals are the last chance to stop the loss. FinCEN lists payments to "offshore digital asset exchanges" among the ways fraudsters make funds hard to recover.

Layered Controls

NIST's identity proofing guideline, SP 800-63A-4 (final, July 2025), is a useful baseline. FinCEN and ENISA add the rest.

Prove the camera is real. NIST says providers shall use controls to increase confidence that media comes from a genuine sensor.
Attest the device. NIST says providers should authenticate capture sensors or use device attestation.
Analyze every capture. Look for manipulation and for signatures of known deepfake tools.
Randomize liveness. Pair unpredictable active challenges with passive checks, as ENISA describes.
Check the document itself. NIST requires document presence checks; FinCEN cites reverse image search and metadata.
Step up on risk. FinCEN points to phishing-resistant MFA and live audio or video verification.
Re-review after the fact. FinCEN says institutions often catch AI documents by re-reviewing account opening files.

Use a dedicated app where you can. ENISA notes an app can check the device is not an emulator and block injected video and audio.

Questions for your KYC vendor

NIST requires providers to test detection against forged media and to share documented false negative rates with relying parties on request.

Ask for those, then:

Which injection routes do you detect: virtual cameras, emulators, hooked apps?
How does detection work in a browser, not just your app?
Does your verdict reach our server directly, or through the user's device?

The last question is easy to miss. If the client app can report "verified" by itself, no liveness model protects you.

Where Guardian Fits

Guardian's fintech pentests attack "the flows that authenticate users, authorize actions and approve withdrawals, function by function".

The web app pentest covers authentication, value-moving flows, business logic and every parameter the app accepts.

The API pentest covers key scoping, per-object authorization, rate limiting and replay, and "the account-recovery paths that bypass all of it".

That is where a KYC verdict meets your platform. For the wider picture, read our guide to crypto exchange security.

Frequently Asked Questions

Can deepfakes bypass KYC at crypto exchanges?

They can get through weak checks. FinCEN says criminals have opened accounts with suspected AI-made identities, and NIST says liveness checks cannot stop every injection attack. Layered controls and monitoring after onboarding reduce the risk.

What is a camera injection attack?

Forged or modified video is inserted between the camera and the verification system, often through a virtual camera, emulator or webcam plugin. The real camera never sees it, so checks on what the camera sees can miss it.

What is the difference between liveness detection and injection detection?

Liveness detection asks whether a live person is in front of the camera. Injection detection asks whether the video came from a genuine camera at all. An injected deepfake can pass the first and fail the second.

What should I ask my KYC vendor?

Ask which injection attacks it detects, how it tests detection against forged media, and for its documented false negative rates, which NIST expects providers to share on request. Confirm the verdict reaches your server directly.

Sources

Test the flows behind your KYC checks

Guardian pentests the authentication, account recovery and withdrawal flows that a passed identity check unlocks.

Talk to Guardian Explore fintech security