Oct 21, 2026

AI-Powered ID and Passport Issuance: How Artificial Intelligence Is Changing Personalisation and Quality Control

Most conversations about AI and identity focus on biometrics at the border or on face matching in banking apps. A quieter change is happening earlier in the chain, in the machines that personalise ID cards and passports. Here AI is not deciding who someone is. It checks that the document produced is correct, consistent and safe to hand over.

This matters most in decentralised issuance. When documents are personalised in fifty offices instead of one bureau, there is no team of specialists inspecting every card. The machine has to check its own work. This guide explains what AI does in personalisation, where it adds value and what to ask before you rely on it.

Why Quality Control Is the Weak Point of Decentralised Issuance

A central bureau has trained inspectors, controlled lighting and sampling procedures. A town hall has a clerk with a queue of citizens. The clerk cannot examine every card under magnification, and should not have to.

Errors that slip through are costly. A faint or shifted portrait can fail automated border checks. A wrong character in the MRZ makes the card unreadable by machines. A card with a laser defect may look acceptable but fail when inspected. Each faulty document means a reissue, a second visit and a loss of trust in the programme. In the worst case, a defect is mistaken for tampering at a border.

AI-based inspection brings the quality control of a central bureau to every desktop machine.

What AI Does in Personalisation

Automated visual inspection

After engraving, a camera captures the card. A trained model compares the result with the expected layout and with what a good card looks like. It checks that the portrait is present, centred, sharp and within the expected contrast range. It checks that every text field is complete and in the right place. It looks for scratches, burns, missing areas and alignment errors.

Rules-based machine vision has done some of this for years. Its limit is rigidity: every check must be programmed in advance. Learned models handle variation better. They can recognise that a portrait is poor even when no single measured value is out of range.

MRZ and data verification

Optical character recognition reads the MRZ and the printed fields. The system confirms that what was engraved matches what was sent, character by character. It recomputes the check digits defined in ICAO Document 9303. It compares the engraved data with the data written to the chip.

Anomaly detection across the fleet

A single defect may be bad luck. A pattern is a problem. When every machine reports inspection results to a central platform, AI can detect drift: a laser losing power, a lens needing cleaning, a batch of card bodies reacting differently, one site producing more rejects than the others. Maintenance moves from reactive to predictive.

Anomaly detection also supports security. A machine that suddenly personalises cards at unusual hours, in unusual volumes or with unusual data patterns deserves attention. AI can flag it for a human to review.

Portrait quality at enrolment

The best engraving cannot rescue a poor photograph. AI tools at the enrolment stage check the portrait before personalisation: pose, expression, lighting, background, sharpness and compliance with ICAO portrait requirements. They can also help detect image manipulation, such as face morphing, where two faces are blended so that one document matches two people.

Operator assistance

AI assistants can guide clerks through procedures, explain error messages and answer questions about configuration. LIZAR’s DRAGO assistant, for example, already answers technical questions on the company’s website. Inside the issuance workflow, the same approach reduces training time and support calls.

What AI Should Not Do

AI in personalisation is a quality and monitoring tool. It should not make identity decisions on its own.

It should not decide whether an applicant is entitled to a document. That remains a legal decision taken by a person.

It should not silently correct data. If the engraved data does not match the request, the card should be rejected and the error logged, not adjusted.

It should not replace security controls. Operator authentication, central authorisation of every job and blank stock control remain essential.

A clear division helps. AI inspects, flags and recommends. People and rules decide.

Benefits in Practice

  • Defect detection. Without AI inspection: Depends on the clerk’s attention. With AI inspection: Every card checked the same way.
  • MRZ and data errors. Without AI inspection: Found later, often at a border. With AI inspection: Found before the card leaves the machine.
  • Machine drift. Without AI inspection: Noticed after many rejects. With AI inspection: Detected early across the fleet.
  • Reissues and second visits. Without AI inspection: Higher. With AI inspection: Lower.
  • Evidence for audits. Without AI inspection: Partial, manual. With AI inspection: Full inspection record per document.
  • Training needs. Without AI inspection: Clerks learn to spot defects. With AI inspection: Clerks act on clear flags.

The inspection record also has value of its own. Each document carries evidence that it was checked and passed. If a card is later questioned, the agency can show the inspection result.

AI and Sovereignty

Identity data is among the most sensitive data a state holds. Before adopting AI in issuance, ask where the models run and where the data goes.

Inspection should run on the machine or inside the agency’s own infrastructure, not in a public cloud outside the jurisdiction. Training data should be controlled, documented and free of personal data where possible. Model updates should be signed, tested and deployed under the agency’s control.

LIZAR designs its AI tools for this context. Inspection runs within the issuance ecosystem. Hardware is designed, engineered and assembled in Europe, and the platform supports centralised, distributed and portable deployment.

The AI Act

The EU AI Act, Regulation (EU) 2024/1689, takes a risk-based approach. Obligations depend on what a system does. Systems used for remote biometric identification of people, for example, face strict rules. A system that inspects the physical quality of a printed card performs a very different function.

The classification of a specific tool depends on its exact purpose and context. Agencies should assess each AI component with their legal team and ask suppliers for documentation: what the model does, how it was trained, how it is tested and how humans stay in control. Good suppliers will have this ready.

How to Evaluate AI Inspection

Ask what the system inspects. Portrait, text fields, MRZ, chip data, security features and physical defects should all be covered.

Ask how it was validated. Request detection and false-reject rates measured on cards similar to yours, not marketing figures.

Ask what happens when it flags a card. The workflow should be clear: reject, log, reissue, and alert if a pattern appears.

Ask where it runs and where the data goes. Local or on-premise processing should be the default for identity data.

Ask how models are updated. Updates should be versioned, signed and tested before deployment across the fleet.

Ask for a live test. Personalise a batch with deliberate defects and see how many the system catches.

Where This Is Going

Three trends are clear. First, issuance is moving closer to citizens, which makes automated quality control a requirement rather than an option. Second, physical and digital identity are converging: the same card that is inspected at personalisation will anchor a digital wallet under eIDAS 2.0. Third, fleets are becoming software-defined, with AI watching the health and security of every unit in real time.

For agencies, the practical step is simple. When you buy personalisation equipment, treat inspection and monitoring as core requirements, not as options.

Frequently Asked Questions

How is AI used in ID card personalisation?

Mainly for quality control. AI inspects each card after engraving, verifies the portrait, data and MRZ, and detects defects. It also monitors fleets of machines to spot drift and unusual activity.

Does AI decide who gets an ID document?

No. Entitlement to a document remains a legal decision taken by people. AI in personalisation checks the quality and consistency of the document produced.

What is AI quality inspection for ID cards?

A camera captures each personalised card and a trained model checks the portrait, text fields, MRZ and physical condition against the expected result. Faulty cards are rejected before they reach the citizen.

Can AI detect face morphing in passport photos?

AI tools can help detect signs of image manipulation, including morphing, at the enrolment stage. They support trained staff and do not replace them.

Is AI in ID issuance covered by the EU AI Act?

The AI Act applies according to the purpose of each system. Biometric identification is treated very differently from inspection of printed output. Each tool should be assessed with legal advice and supplier documentation.

Where should AI inspection run?

On the machine or within the agency’s own infrastructure, so that identity data does not leave the jurisdiction. Model updates should be signed and controlled by the agency.

What are the benefits of AI inspection in decentralised issuance?

Every card is checked to the same standard, errors are caught before delivery, machine problems are detected early and each document has an inspection record for audits.

Related reading

Sources

Share this resource

RELATED NEWS

Secure ID Cards: A Practical Guide to Materials, Security Features and Issuance

28 OCT

Secure ID Cards: A Practical Guide to Materials, Security Features and Issuance

What makes an ID card secure: card materials, laser engraving, CLI and MLI, chips, MRZ and the checks a government or company should run before buying.

Learn More
CLI and MLI Explained: How Changeable and Multiple Laser Images Protect ID Cards

21 OCT

CLI and MLI Explained: How Changeable and Multiple Laser Images Protect ID Cards

What CLI and MLI laser images are, how they are engraved through a lens structure, how to inspect them and what card bodies and equipment they need.

Learn More

21 OCT

eIDAS 2.0 and Physical ID Cards: Why the Digital Wallet Makes the Card More Important, Not Less

Will the EU Digital Identity Wallet replace ID cards? What eIDAS 2.0 requires, the 2026 and 2027 deadlines, and why the physical card anchors the wallet.

Learn More
More News

Secure credentials · Controlled issuance · Trusted infrastructure

Let's build the future of secure identity.

DRAGO
Assistant

👋 Hi! I'm DRAGO.

How can I help you today?