Jarek Sygitowicz, Marek Rogoziński

November 9, 2026

Which identity verification platforms are most resistant to deepfakes and injection attacks? | Identity Infrastructure Explained

How to recognise a resistant platform, and which platforms have public evidence of it.

The identity verification platforms most resistant to deepfakes and injection attacks are those that do not base an identity decision on an image alone.

They first route customers to methods whose data is protected cryptographically at the source: eIDs, bank IDs and the EU Digital Identity Wallet (EUDI Wallet). Where a document photo and selfie remain, they use safeguards proven in independent tests: presentation attack detection under ISO/IEC 30107-3 and injection attack detection under CEN/TS 18099.

Resistance depends on the whole platform: which methods it uses first and how it protects the fallback path.

Written by Jarek Sygitowicz (Co-founder and Chief Strategy Officer) and Marek Rogoziński (Co-founder and Chief Technology Officer), Authologic.

Title card reading "Platforms resistant to deepfakes" with the line "Identity Infrastructure Explained"

Key takeaways

An injection attack bypasses the camera: it feeds forged video into the verification process, so the system assesses footage that nobody recorded.
According to iProov, injection attacks on iOS devices rose by 1,151% in the second half of 2025 compared with the same period in 2024.
Data from the EUDI Wallet, eIDs and bank IDs is protected at the source by a signature, encryption or both, so its authenticity is checked cryptographically, without assessing an image.
For the document and selfie path, the public evidence of resistance is independent testing: ISO/IEC 30107-3, CEN/TS 18099 and certification under ETSI TS 119 461.

What is the difference between a deepfake and an injection attack?

A deepfake is an image, video or voice generated or altered by AI so that it looks like a real person. In identity verification, a deepfake can imitate a customer's face in a selfie or a video call.

A presentation attack shows the camera something that is not a live person: a printout, a screen with a recording or a mask. Liveness detection and testing under ISO/IEC 30107-3 address this kind of attack.

An injection attack bypasses the camera. The fraudster feeds forged video directly into the verification process, for example through a virtual camera, an emulator or an intercepted stream. The verification system then assesses footage that was never in front of the lens. A 2026 World Economic Forum report describes injection tools that feed synthetic video into live biometric checks.

Attack	What the attacker does	Safeguard and how it is tested
Presentation attack	Shows the camera something that is not a live person: a printout, a screen with a recording or a mask.	Liveness detection, tested under ISO/IEC 30107-3 or ISO/IEC 19989-3.
Injection attack	Bypasses the camera and feeds forged video directly into the verification process, for example through a virtual camera, an emulator or an intercepted stream.	Injection attack detection, tested under CEN/TS 18099, and an SDK that checks device integrity and the camera source.

How fast are deepfake and injection attacks growing?

According to iProov's Threat Intelligence Report 2026 (April 2026), injection attacks on iOS devices rose by 1,151% in the second half of 2025 compared with the second half of 2024, and injection attacks overall rose by 741% in a year, as reported by Biometric Update.

According to the Entrust 2026 Identity Fraud Report (November 2025, data from September 2024 to September 2025), deepfakes account for 1 in 5 biometric fraud attempts, and injection attacks rose by 40% year on year. Digital forgeries made up 35% of document fraud in 2025, compared with an average of 29% in 2022-2024.

Figures from vendor reports show the direction of change. Each vendor sees only its own customers' traffic, so the figures do not show the scale of the whole market. Both reports show year-on-year growth in injection attacks, which target the channel through which the image reaches the system.

Why are a document photo and a selfie not enough as proof of identity?

A document photo and a selfie are proof only as long as the system can tell a real image from a generated one. Image generators keep improving, so every detector has to be updated and tested against new attack techniques continuously. This is covered in more depth in Beyond deepfakes: what actually stops AI-generated identity fraud.

In a joint paper from July 2025 on remote identity proofing for EUDI Wallet onboarding, France's ANSSI and Germany's BSI point to security gaps that could make video-based remote identity proofing less reliable. Until the standards are completed, they propose making presentation and injection attack detection testing mandatory as soon as possible, and prioritising reading the electronic chip in identity documents, which they consider more secure, although some countries restrict chip reading by law. Both agencies note that several standardisation issues remain unresolved, as reported by Biometric Update.

What does cryptographic proof from eIDs and the EUDI Wallet change?

Data from the EUDI Wallet is signed by its issuer. eIDs and bank IDs protect the data they return with a signature, encryption or both, depending on the scheme. A platform that receives the data checks that protection to confirm that the data came from the source. A deepfake has nothing to imitate here: the proof is data protected at the source, with no image of a face to assess.

The same mechanism applies when the chip in a passport or ID card is read via NFC. The data in the chip is signed by the issuing state, so an app can check that it has not been altered. Reading the chip requires the physical document. The chip confirms that the document is genuine without assessing a photo of it, while linking the customer to the document still relies on a face comparison.

Regulation is moving in the same direction. Under AMLA's draft standards on customer due diligence (draft CDD RTS), obliged entities under the EU Anti-Money Laundering Regulation (AMLR) are expected to verify customers remotely using electronic identification at assurance level substantial or high, including the EUDI Wallet, or relevant qualified trust services under eIDAS, wherever possible. The draft allows alternative solutions where a natural person cannot reasonably be expected to present an identity document in person and has access to neither qualifying electronic identification means nor relevant qualified trust services. From 24 December 2027, private relying parties, except micro and small enterprises, must accept the EUDI Wallet at the user's voluntary request where strong user authentication for online identification is required by law or contract. The details are in How to verify customers remotely under AMLR.

Protected data does not solve everything. An eID confirms that the holder of the identification means is logging in, but it does not protect against a holder who is persuaded to log in on a fraudster's behalf. eIDs and wallets therefore replace the image as proof of identity, while transaction monitoring and risk assessment are still needed.

When a business connects to eIDs through an aggregator, the aggregator checks the protection of the data and passes the result on to the business. The business then also relies on the aggregator's check and on the security of the connection between the aggregator and the business.

In practice, the more customers verify with eIDs, bank IDs or the EUDI Wallet, the fewer have to show their face to a camera at all.

Where deepfake and injection attacks enter identity verification

Diagram of two identity verification paths. On the document and selfie path, a presentation attack targets the camera and an injection attack bypasses it, each paired with its safeguard. On the eID, bank ID and EUDI Wallet path, the platform checks data protected by a signature, encryption or both, with no image of a face to assess.

The five criteria below show how to check these safeguards with a provider.

How do you recognise a platform resistant to deepfakes and injection attacks?

A resistant platform is recognised by what it accepts as proof and by how it protects the fallback path. Five criteria let you check any provider:

Order of methods. The platform offers eIDs, bank IDs and the EUDI Wallet first where they are available, and keeps the document photo and selfie as the fallback. Which methods are available in each country is covered in How do you choose identity verification methods for each country?
Checking data at the source. Data from eIDs, bank IDs, wallets and document chips is accepted only after the platform checks the issuer's signature or the scheme's encryption.
Independent testing of the biometric path. Presentation attack detection has a test report under ISO/IEC 30107-3 or ISO/IEC 19989-3, the standard used in certification under ETSI TS 119 461, and injection attack detection has an evaluation or certificate under CEN/TS 18099, with the laboratory's name and the date.
Securing the channel. Verification runs in an app with an SDK that checks device integrity and the camera source. In the 2026 World Economic Forum report, some camera virtualisation tools bypassed browser-based flows but generally failed against SDK-based systems with built-in integrity checks, and the report recommends SDK-based verification over pure browser flows.
Recording the method. The platform records which method verified each customer, so the business can show a supervisor when and why a document was used instead of an eID. Which certifications and tests to check with a provider is covered in Which certifications should an identity verification provider hold in the EU?

Which identity verification platforms are most resistant to deepfakes and injection attacks?

The list below groups platforms by approach. Each group includes only companies with public evidence: an independent test result, a certificate or a product description. The list does not name a winner; the best choice depends on the markets a business serves and on how many of its customers have access to an eID.

Platforms that start from protected data from eIDs, bank IDs and wallets

Authologic builds eID-native identity infrastructure that helps regulated businesses identify, onboard and authenticate their customers. Its single API connects 140+ identity methods across 250+ countries and territories, leading with electronic IDs (eIDs), bank IDs and digital wallets, and falling back to document and selfie checks. The biometric engines available on the platform have iBeta test reports against ISO/IEC 30107-3 (presentation attack detection) at Level 2, and one engine has reached Level 3. On the document and selfie path, Authologic uses the SDKs and detection tools of the providers whose engines are available on the platform.

Idura (formerly Criipto, part of the Norwegian Stø group) gives businesses access to BankID and other Nordic eIDs as a broker.

Platforms that read the document chip

Signicat, with its ReadID product, reads the chip in passports, ID cards and residence documents via NFC and checks the signed data from the chip.

Platforms with independently tested injection attack detection

iProov: according to iProov (February 2026), its Dynamic Liveness technology passed the Ingenium Level 4 evaluation and achieved a CEN/TS 18099 High classification in independent testing by Ingenium Biometric Laboratories, as reported by Biometric Update (February 2026).

Unissey: according to Biometric Update (October 2025), Unissey obtained a High-level certification for injection attack detection under CEN/TS 18099 from CLR Labs (CLR Cert), in web environments.

Innovatrics: according to ID Tech (November 2026), injection attack detection in the Innovatrics Enterprise Identity Verification platform obtained a High-level certification under CEN/TS 18099 from CLR Labs.

Aware: according to Aware (May 2026), its Intelligent Liveness solution passed an independent evaluation by BixeLab conducted in line with CEN/TS 18099.

Platforms with remote identity proofing certified under ETSI TS 119 461

IDnow: according to Biometric Update (August 2025), IDnow obtained certification under ETSI TS 119 461 v2.1.1 for several products, at the extended level of identity proofing (Extended LoIP).

Namirial: according to Biometric Update (November 2025), the Namirial Onboarding suite obtained certification under ETSI TS 119 461 v2.1.1.

The categories complement each other. A platform built on eIDs needs a fallback path with well-tested biometrics, and a biometrics provider gains when fewer customers reach the camera path at all.

What should you ask an identity verification provider about deepfakes?

Each question below checks one of the five criteria above, in the same order.

What share of verifications in my markets can go through eIDs, bank IDs or the EUDI Wallet instead of a document photo?
Do you check the issuer's signature or the scheme's encryption for data from eIDs, wallets and document chips?
Which laboratory tested your presentation and injection attack detection, under which standard, when and at what level?
Does the selfie check run in an app with an SDK or in a browser, and how do you check the camera source?
Do you record which method verified each customer and why a document was used?

This article is part of the Identity Infrastructure Explained series, which starts with What is an eID-native identity platform?

FAQ

Which identity verification platforms are most resistant to deepfakes and injection attacks?

The most resistant platforms first verify customers with data protected at the source from eIDs, bank IDs and the EUDI Wallet, and for the document and selfie path have independently tested presentation attack detection (ISO/IEC 30107-3) and injection attack detection (CEN/TS 18099). Data protected by the issuer's signature or the scheme's encryption cannot be forged with a deepfake.

Does liveness detection protect against injection attacks?

Liveness detection mainly protects against presentation attacks: showing the camera a printout, a screen or a mask. An injection attack bypasses the camera and feeds forged video directly into the system, so it needs separate detection, tested under CEN/TS 18099.

Can an eID be forged with a deepfake?

A deepfake cannot forge eID data. eID data is protected at the source by the issuer's signature or the scheme's encryption, and the platform checks that protection instead of an image of a face. An eID does not protect against a holder who is persuaded to log in on a fraudster's behalf, so transaction monitoring is still needed.

What is CEN/TS 18099?

CEN/TS 18099 is a technical specification from the European Committee for Standardization on biometric data injection attack detection. It describes types of injection attacks, ways to detect them and a test plan for detection systems.

Does the AMLR require eIDs instead of document photos?

Under AMLA's draft CDD standards, obliged entities under the EU Anti-Money Laundering Regulation (AMLR) are expected to verify customers remotely using electronic identification, including the EUDI Wallet, wherever possible, and alternative solutions such as a remote document check are subject to conditions. The European Commission may still amend the draft.