Real-time remote biometric identification

Capture, comparison and identification occur instantaneously, near-instantaneously or without significant delay using live or near-live material.

In Recital (12), a machine-based system distinguished from simpler software by characteristics including the capability to infer, producing outputs that can influence physical or virtual environments.

The process of obtaining outputs such as predictions, content, recommendations or decisions, and the capability to derive models or algorithms, or both, from inputs or data.

Core Concepts, Reach and the Risk-Based Framework — The EU Artificial Intelligence Act: A Section-by-Section Deep Dive into Regulation (EU) 2024/1689

Reading the Recitals: A Conceptual Map

Recitals as explanation

Recitals (12) to (27) explain the Regulation's concepts, reach, exclusions and policy logic. Their wording matters: a recital's should remains a should.

A reading sequence

Start with the system: is there AI? Then identify the deployer, biometric function, location and territorial connection before considering exclusions.

The final framework

The text culminates in Recitals (26) and (27): rules should be risk-based, while non-binding ethics guidelines inform trustworthy, human-centric AI.

Recital (12): What Makes a System an AI System?

The key distinction

Recital (12) separates AI systems from simpler software and rules defined solely by people to automatically execute operations.

Inference

"A key characteristic of AI systems is their capability to infer." Outputs can be predictions, content, recommendations or decisions influencing physical or virtual environments.

Beyond data processing

Inference can derive models or algorithms from inputs or data. It transcends basic processing through learning, reasoning or modelling.

System design and operation

AI may have explicit or implicit objectives, varying autonomy, post-deployment self-learning, and may be standalone, embedded or non-embedded in a product.

Recitals (13) to (18): Deployer and Biometric Concepts

Deployer

A deployer is "any natural or legal person, including a public authority, agency or other body, using an AI system under its authority, except where the AI system is used in the course of a personal non-professional activity."

Identification versus verification

Biometric identification establishes identity by comparing a person's biometric data against stored data in a reference database. Verification or authentication for access has a different, limited sole purpose.

Categorisation

Biometric categorisation assigns people to categories using biometric data. A commercial feature may be ancillary only where it is intrinsically linked, technically inseparable and not a circumvention device.

Emotion recognition

Emotion recognition identifies or infers emotions or intentions from biometric data. It does not include pain or fatigue detection, or mere visible expressions unless used to infer emotion.

Worked Examples: Identify the Biometric Function

Phone unlock

Matching one claimant to an enrolled template solely to unlock a device is biometric verification or authentication, not biometric identification.

Stadium search

Scanning multiple people at a distance and comparing faces against a reference database to establish identity fits remote biometric identification.

Marketplace filter

A glasses-preview filter may be ancillary where it is technically tied to the marketplace's principal service and helps the consumer make a purchase decision.

Fatigue alert

Detecting fatigue to prevent accidents is a physical-state use. Recital (18) says physical states such as pain or fatigue are not emotion recognition.

Recitals (17) and (19): Remote Identification and Public Space

Functional definition

Remote biometric identification concerns identification without active involvement, typically at a distance, through comparison against a biometric reference database.

Real-time and post

Real-time capture, comparison and identification occur with no significant delay. Post systems analyse material captured earlier after a significant delay.

Physical and undetermined

A publicly accessible space is "any physical space that is accessible to an undetermined number of natural persons". Public or private ownership does not decide the issue.

Boundaries matter

Ticketing or registration can still allow public access. Offices, employee-only workplaces, prisons, border control and online spaces are not publicly accessible under this recital.

Boundary Test: Is It Public and Is It Remote?

Apply Recitals (17) and (19)

For each setting, decide two things:

Does it appear to be a publicly accessible physical space under Recital (19)?
If a system compares facial images with a reference database without the person's active involvement, would the activity have the functional features of remote biometric identification under Recital (17)?

A ticketed concert hall: A ticket condition can be fulfilled by an undetermined number of persons. It can therefore still be publicly accessible.
An employee-only laboratory behind badge access and restriction signs: It is intended for defined employees and service providers. It is not publicly accessible merely because a door was accidentally left open.
The corridor of a private apartment building leading to a doctor's office: The recital notes that a space can contain both publicly accessible and non-publicly accessible areas. Assess the exact area and access conditions.
A public social-media livestream: It may be publicly viewable online, but online spaces are not physical spaces and are therefore outside this recital's definition of publicly accessible space.

Now add timing. A live camera matching faces as people pass through a station is real-time if there is no significant delay. Searching stored station footage days later is a post system. The labels depend on how the system operates, not on what its vendor calls it.

Recitals (21) to (23): Territorial Reach and Third-Country Outputs

Non-discriminatory reach

Recital (21) says rules should apply to providers without discrimination whether they are established in the Union or a third country, and to deployers established in the Union.

The output connection

"this Regulation should also apply to providers and deployers of AI systems that are established in a third country, to the extent the output produced by those systems is intended to be used in the Union."

Cooperation framework

A narrow recital-level exception concerns third-country public authorities and international organisations acting in qualifying law-enforcement or judicial-cooperation arrangements with adequate safeguards.

Accountability remains

Recipient national authorities and Union bodies using outputs in the Union remain accountable for ensuring that use complies with Union law. Union institutions may themselves be providers or deployers.

Worked Examples: Scope, Exclusions and Return to Scope

Union-bound output

A third-country system can fall within the recital's reach where its output is intended to be used in the Union, even when the system is not used there.

Purpose can change scope

A military, defence or national-security system is excluded for those purposes. Civilian, humanitarian, law-enforcement or public-security reuse brings it into scope.

Research is not a permanent shield

Sole-purpose scientific R&D is excluded, but an in-scope system resulting from R&D must comply when placed on the market or put into service.

Recitals (20), (26) and (27): Literacy, Risk and Ethics

AI literacy

Recital (20) says AI literacy should help providers, deployers and affected persons make informed decisions, with knowledge adapted to their context.

External currency note

PE-CONS 30/26 of July 8, 2026 is signed pending publication and revises Article 4 on AI literacy. It is not in force as of July 22, 2026.

Risk calibration

"That approach should tailor the type and content of such rules to the intensity and scope of the risks that AI systems can generate."

Three responses

Recital (26) describes prohibiting certain unacceptable practices, requirements and obligations for high-risk systems, and transparency obligations for certain systems.

Flashcards: Core Terms and Distinctions

Flip each card, then explain the distinction aloud in your own words.

Inference: A core AI characteristic. It concerns obtaining outputs such as predictions, content, recommendations or decisions, and can include deriving models or algorithms from inputs or data.
Deployer: Any natural or legal person, including a public authority, agency or other body, using an AI system under its authority, except personal non-professional use.
Biometric identification: Automated recognition to establish identity by comparing a person's biometric data with stored biometric data in a reference database.
Biometric verification: Authentication whose sole purpose is confirming that a specific person is who they claim to be for access to a service, device or premises.
Real-time remote biometric identification: Capture, comparison and identification occur instantaneously, near-instantaneously or without significant delay using live or near-live material.
Post remote biometric identification: Biometric data was captured earlier; comparison and identification occur only after a significant delay.
Publicly accessible space: Any physical space that is accessible to an undetermined number of natural persons. Online spaces are not covered by this definition.
Risk-based approach: An approach that should tailor the type and content of rules to the intensity and scope of risks AI systems can generate.

Knowledge Check 1: Inference and Biometrics

Choose the answer that best follows Recitals (12), (15), (17) and (18).

Which scenario is excluded from the recital's definition of biometric identification because it is biometric verification or authentication?

A station camera compares every passerby's face with a police reference database.
A phone confirms that its enrolled owner is the person claiming access before unlocking.
A system sorts travellers by estimated age using facial images.
A system analyses archived CCTV footage to identify an unknown person.

Show Answer

Answer: B) A phone confirms that its enrolled owner is the person claiming access before unlocking.

Recitals (15) and (17) exclude systems whose sole purpose is confirming that a specific person is who they claim to be for access to a service, unlocking a device or security access to premises. The other scenarios involve identification, categorisation or a different function.

Knowledge Check 2: Reach, Exclusions and Ethics

Choose the answer that most precisely reflects the source text.

A third-country provider operates an AI system outside the Union, but its output is intended for use by a contracting operator in the Union. What does Recital (22) say?

The Regulation should never apply because the system operates outside the Union.
The Regulation should apply only if the provider is a Union company.
The Regulation should also apply to third-country providers and deployers to the extent the output is intended to be used in the Union.
The Regulation applies only if the AI system was physically placed on the Union market.