What SP 800-63A-4 Asks of Remote Identity Proofing

You are building or buying a remote onboarding flow, and you want to know what a serious technical baseline for identity proofing looks like when nobody stands next to the applicant. This guide reads Special Publication 800-63A-4, the identity proofing volume of the United States Digital Identity Guidelines, and answers one question: what does it require of evidence, validation and verification when the person is remote, and where do injection attacks and forged media enter?

The publication is a US federal guideline, not a law of any other country, and it speaks to US federal services. Its requirements are concrete and measurable, which makes it a useful reference for any team that designs remote proofing. Everything below is taken from the publication itself, and the requirements are described in its own terms: "shall" marks a requirement, "should" a recommendation.

Where the publication sits

The US National Institute of Standards and Technology released the final version of Revision 4 of its Digital Identity Guidelines in July 2025. Its landing page describes an almost four-year process that included two public drafts and nearly 6,000 individual comments. Among the substantial changes the page lists are expanded fraud requirements for identity proofing, a restructuring of identity proofing controls "to better define roles and types of identity proofing", and new controls for injection attacks and forged media such as deepfakes.

Volume A, the subject of this article, covers identity proofing and enrollment. Its abstract describes the core exchange: an applicant provides evidence to a credential service provider, which reliably identifies them. In the rest of this article the credential service provider is simply called the provider.

The three steps

The publication presents a three-step process. Identity resolution collects evidence and attributes to establish that the applicant is a unique, real person within the population the provider serves. Identity validation confirms the authenticity, accuracy and validity of that evidence. Identity verification confirms that the applicant presenting the evidence is the same individual to whom it was issued. After the three steps the provider enrolls the applicant into a subscriber account.

For a remote flow the example workflow is easy to recognise. The provider collects one or two pieces of evidence of the right strength. It checks that the evidence is authentic, accurate and valid, and checks the attributes against authoritative or credible sources. It then applies an approved verification method to confirm that the applicant is the genuine owner of the evidence, and sends a notification of proofing to a validated address.

Four proofing types

The publication defines proofing types by two factors, where the proofing takes place and whether an agent of the provider attends. Remote unattended proofing is completely automated, and the provider does not control the location or the devices. Remote attended proofing runs through a secure video session with a proofing agent. On-site unattended and on-site attended proofing happen in a controlled physical setting. A provider may offer several types, and it must design each option so that the options give comparable assurance in aggregate.

Assurance levels in one paragraph

Three identity assurance levels are defined, each building on the previous one. Level 1 supports the real-world existence of the claimed identity and is designed to limit highly scalable attacks such as automated enrollment, and to protect against synthetic identities. Level 2 requires additional evidence and more rigorous validation and verification, and is designed to limit scaled and targeted attacks and to protect against basic evidence falsification, evidence theft and social engineering. Levels 1 and 2 can be delivered remotely or on site, with or without an attending agent. Level 3 adds a trained agent, an on-site attended session and the collection of at least one biometric characteristic.

Evidence: three strengths

Evidence strength depends on how rigorous the issuing process was, how well the evidence can be validated, and whether it can support a verification method. The publication names three strengths, rising from fair to strong to a top tier.

Fair evidence comes from an issuer that followed formal procedures to confirm the identity, is delivered to the person it concerns, carries the name and a reference number, a facial image or enough attributes to be unique, and has security features that make it difficult to reproduce.
Strong evidence additionally rests on written issuing procedures under recurring oversight by regulatory or publicly accountable institutions, and contains a reference number or other attributes that uniquely identify the person, together with a facial image or other biometric characteristic.
Top-tier evidence adds cryptographic protection that can be validated through the issuer's digital signature, and an attended enrollment that confirmed the person's physical existence.

At the first level the provider collects one piece of evidence: fair evidence that can be digitally validated or that includes a facial portrait or other biometric, or strong evidence, or top-tier evidence.

Validation: is the document genuine?

Validation examines the evidence to confirm that it is authentic (not forged or altered), accurate and valid. The acceptable methods are visual and tactile inspection by trained personnel on site, visual inspection by trained personnel for remote proofing, automated document validation, and cryptographic verification of digital evidence.

For automated validation of physical documents the publication sets measurable performance targets. The document false acceptance rate, meaning the share of fraudulent documents the system judged valid, has to be 0.1 or less. The document false rejection rate, meaning the share of genuine documents judged invalid, has to be 0.1 or less. Further requirements follow for such systems. If a machine readable zone or barcode is present, the system compares its data with the printed data for consistency. The provider uses live capture and a document presence check, which the publication also calls document liveness, so that the image captured in the session is not a manipulated digital copy. The provider assesses performance in conditions similar to real operation across all evidence types it accepts, and publishes the results. It should also have the technology tested periodically by independent entities such as laboratories or research institutions.

Verification: is it the owner?

The provider has to verify the applicant's ownership of one piece of evidence by one of several methods: returning a confirmation code sent to a validated address, returning a micro-transaction value, completing an authentication protocol at a sufficient level, an automated comparison of the applicant's facial image with the portrait on the evidence, or a visual comparison by a trained agent in an attended or asynchronous session. Automated comparison of a live sample with the image on or stored with the evidence is also listed.

Biometric thresholds

Where biometrics are used, the publication gets specific. For one-to-one comparison it sets a false match rate of 1:10,000 or better and a false non-match rate of 1:100 or better. For one-to-many search used in resolution or deduplication, the false positive identification rate has to be 1:1,000 or better, and an enrollment cannot be declined on an automated search result alone, without a manual review.

Performance for applicants of different demographic types may be no more than 25 % worse than the performance for the overall population, measured with a single fixed threshold. For example, if the false non-match rate overall is 0.006, a demographic group cannot exceed 0.0075.

For remote collection the provider has to run presentation attack detection that meets an impostor attack presentation accept rate below 0.07, and the tests must conform to the 2023 edition of part 3 of the 30107 series of standards from the International Organization for Standardization and the International Electrotechnical Commission. The provider also has to make the results of its biometric algorithm and system tests publicly available, which it may do in summary form.

Alongside the technical thresholds, the publication asks for explicit, informed consent from all applicants for the collection and use of biometrics, and for a stored record of that consent. It asks for clear public information about every use of biometrics, a documented and publicly available deletion process, and a default retention period consistent with the regulations of the regions and sectors served. Deleting all of a subscriber's biometric information on request is a recommendation, and a provider that does not support it has to document its justification publicly.

Injection and forged media

The section on digital injection prevention and forged media detection describes the newest class of attack. Many attacks on attended and unattended remote proofing pair digital injection with widely available generative AI tools to create or modify images and video of applicants and documents. Injection attacks insert modified or forged media between the capture point and the component that performs the comparison. The publication states that a biometric comparison performed with a captured sample does not prevent these attacks. Live document capture and presentation attack detection make injection of viable forged media more challenging, but "even these mechanisms are not sufficient to address all possible cases".

For all remote proofing the provider is required to do three things. It implements technical controls to increase confidence that media come from a genuine sensor, for example by detecting a virtual camera, a device emulator or a jailbroken device. It analyses all submitted media for signs of modification, manipulation, tampering or forgery, tests those algorithms against available attack artifacts, and documents the kinds of artifacts tested and the corresponding false negative rates. It uses only authenticated protected channels. The publication also recommends, as a should and not a shall, that algorithmic analysis of media and automated decisioning be augmented by manual review to address detection errors. Further recommendations add what the publication calls a passive means of detecting forged media in all capture scenarios, authenticating capture sensors or using device attestation, and analysing media for signatures of known generative AI and deepfake tools. For attended sessions the provider has to train agents to look for manipulated media, such as high latency or synchronisation problems, and to introduce random human-in-the-loop cues, such as asking the user to move an object between the sensor and the face.

A design checklist drawn from the text

These points restate the requirements above as questions a product owner can put to any proofing design.

Which proofing type are you offering, and does each option give comparable assurance?
Which evidence strength do you accept for the assurance level you target, and how is each piece validated?
Do you have measured document false acceptance and false rejection rates for your own document mix, and are they published?
Do you know your comparison thresholds and the spread across demographic groups?
How do you detect virtual cameras and emulators, and how do you test your forged-media analysis?
What do applicants consent to, how long is biometric data retained, and how can it be deleted?

How NeuroVision fits

A verification flow built on NeuroVision KYC places these steps in one session: document, selfie, liveness and face match, with the scenario assembled in an editor. The document step detects, classifies and reads IDs, including printed and handwritten text and the machine readable zone with its checksum. Document anti-fraud checks, covering screen photos, photocopies, printouts, glued photos, drawn-on elements, erasures, edited images and logical data errors, are included in every package. The liveness step is active: the person moves the face closer through two ovals. Deepfake and face-swap detection works inside the selfie and liveness steps. Erasure endpoints and a callback that can omit names, document data and images, keeping the session identifier, which links to the session, give the integrating team controls for deletion and data minimisation, and retention in days is a setting of each scenario.

To see how these steps look in your own scenario, contact the team.

Sources

All sources were fetched on 27 October 2026.

Digital Identity Guidelines, Special Publication 800-63A-4, Identity Proofing and Enrollment, US National Institute of Standards and Technology.
Digital Identity Guidelines, Special Publication 800-63-4, landing page, US National Institute of Standards and Technology.