Article 49 requires providers of high-risk AI systems referred to in Article 6(2) (Annex III) to register themselves and their systems in the EU database under Article 71 before placing the system on the market or putting it into service. Certain Annex I product AI systems register only when placed on the market or put into service without a third-party conformity assessment under that product law. Public-authority deployers must register use of specified high-risk systems. Submitted data follow Annex VIII and must stay complete, correct, and up to date after initial filing, in line with technical documentation, conformity assessment, and the EU declaration of conformity. Most high-risk duties, including database registration, apply from 2 August 2026.
Compliance Guide

EU AI Act Article 49 Registration Database

When and how providers and public-authority deployers must register high-risk AI systems, what Annex VIII requires, and how to keep records accurate across the lifecycle.

Article 49 at a glance

Who registers

Providers of Annex III high-risk systems; limited Annex I cases; public-authority deployers for specified uses.

When

Before placing on the market or putting into service; maintain accuracy thereafter.

What data

Annex VIII fields tied to identity, purpose, conformity, and technical documentation.

Evidence link

Registration must align with Annex IV docs, Article 43 assessment, and EU declaration of conformity.

Scope: which systems and actors fall under Article 49

Article 49 is a market-access control for high-risk AI, not a general AI inventory rule. Providers must register high-risk AI systems referred to in Article 6(2), meaning systems listed in Annex III, in the EU database established by Article 71 before placing the system on the market or putting it into service. That registration obligation sits alongside provider duties in Articles 16 to 27 and depends on a defensible high-risk classification under Article 6.

Providers of high-risk AI systems covered by Annex I product legislation (Article 6(1)) register in the EU database only where those systems are placed on the market or put into service without undergoing a third-party conformity assessment under that product law, subject to the conditions in Article 49. Prohibited practices under Article 5 and general-purpose AI model rules are outside Article 49 registration scope.

Deployers that are public authorities, Union institutions, bodies, offices, or agencies, or that act on their behalf, must register use of specified high-risk AI systems under Article 49(3). Private enterprise deployers should still determine whether they are also providers in law (for example through substantial modification or own-name placing on the market), because role determination drives who files what.

For high-risk AI systems in Annex III points 1, 6, and 7 in law enforcement, migration, asylum, and border control management, only a limited non-sensitive subset of Annex VIII data is made publicly accessible. Other data remain available to the Commission and national authorities. Enterprises operating in those domains need dual handling: internal records that are complete for audit, and controlled public-facing fields that match statutory visibility limits.

The AI Act entered into force on 1 August 2024. Obligations apply on a phased timeline. Most high-risk system duties, including database registration, apply from 2 August 2026 unless a specific earlier or later date applies to a given requirement.

Registration scope by actor and system type
Actor / system Registration duty
Provider, Annex III (Art. 6(2)) Register self and system in EU database before market placement or putting into service
Provider, Annex I product AI (Art. 6(1)) Register only when placed on the market or put into service without third-party conformity assessment under that product law
Public-authority deployer (Art. 49(3)) Register use of specified high-risk systems
Article 5 prohibited practices / GPAI models Outside Article 49 registration scope

What must be submitted and how it connects to conformity

Information submitted on registration is set out in Annex VIII, Sections A to C. In practical terms, fields cover provider identity and contact data, system identification and versioning, intended purpose, conformity-related information, and deployer registration data where Article 49(3) applies. These fields map directly to elements already required in Annex IV technical documentation: system identity, intended purpose, version control, conformity route, and provider particulars.

Registration is administratively linked to completion of conformity assessment (Article 43), drawing up the EU declaration of conformity (Article 47), and CE marking where required (Article 48). Registration should not be treated as a standalone form. The database entry must be consistent with the declaration of conformity and with technical documentation kept up to date. If declared purpose, system identity, or conformity route diverge from Annex IV records, market-surveillance review will treat that inconsistency as a compliance failure, not a clerical mismatch.

Article 71 establishes the EU database, managed by the Commission, with public accessibility of registered information except where Article 49 restricts visibility for sensitive law-enforcement and migration systems. Operational detail of the Commission-operated interface, APIs, and submission workflows may continue to evolve in implementing guidance. Enterprise programs should therefore store canonical registration payloads and submission evidence internally, independent of any single portal workflow.

After initial registration, providers must ensure that registration information remains complete, correct, and up to date. That continuing duty connects to post-market monitoring and to substantial modification analysis. A change that alters intended purpose, risk profile, or the system as placed on the market can require registration updates and, where applicable, renewed conformity assessment.

Consistency check

Treat Annex VIII as a published view of Annex IV, Article 43 outcomes, and the EU declaration of conformity. Any gap between those artifacts and the database entry is a compliance risk.

Governance processes that keep registration accurate

Article 49 is an ongoing accuracy and lifecycle control duty, not a one-time filing at go-live. Enterprises need a system inventory keyed to Annex III and Article 6 classification, with unique system and version identifiers reused in technical documentation, declarations of conformity, and database submissions. Without stable identity, registration becomes a snapshot that drifts from the system actually operated.

Accountable ownership should be explicit for every registrable system: legal provider entity, technical owner, and compliance approver. Dual-control validation before submission reduces the risk that marketing descriptions, engineering nicknames, or incomplete intended-purpose statements enter the official record. Registration records should link in a controlled configuration store to technical documentation repositories, conformity-assessment outcomes, and EU declarations of conformity so reviewers can trace each Annex VIII field to an approved source artifact.

Change management must treat substantial modification and intended-purpose changes as compliance events. Metadata for model updates, capability expansion, deployment environment changes, and new tool integrations should trigger a registration accuracy review. Where the change is substantial under the Act, renewed conformity work and database updates are part of the same control path, not sequential afterthoughts.

Cross-functional operating rhythm is required among legal, AI governance, security, engineering, and audit. Legal owns role determination and exclusion rationales, such as why a given Annex I product path does not require EU database registration after third-party assessment. Governance owns the inventory and release gate. Engineering owns version identity and environment controls. Audit owns evidence retention for market surveillance: classification rationale, Annex IV packages, declarations of conformity, registration submission receipts, and change logs.

For law-enforcement and migration-related Annex III systems, separate public versus restricted data handling so internal records exceed what is publicly listed while still meeting authority-access needs. Document the legal basis for restricted visibility rather than relying on informal redaction.

Enterprise registrability workflow

Use a controlled path from classification through release so registration stays aligned with the system you operate.

  1. Classify and inventory

    Map each system against Article 6(2)/Annex III and relevant Article 6(1) cases. Assign owner and stable version IDs shared across docs and filings.

  2. Determine legal role

    Document provider versus deployer status, including any public-authority deployer registration duty under Article 49(3).

  3. Map Annex VIII to source artifacts

    Link every field to Annex IV technical documentation, conformity outcomes, and the EU declaration of conformity before submission.

  4. Enforce a release gate

    Block market placement or putting into service until registration and conformity status are valid for that version.

  5. Trigger updates on change

    Treat substantial modification and purpose change as events that force registration re-validation and, where needed, renewed conformity work.

  6. Retain evidence and verify runtime alignment

    Keep evidence packs and use runtime identity, permissions, and tool-call logs to confirm continued purpose alignment.

Runtime controls as registration evidence

Registration describes the system that may be placed on the market or put into service. Runtime governance shows whether production operation stays inside that description. Version IDs, deployment environment tags, and intended-purpose constraints in runtime configuration supply evidence that registered descriptions match systems actually run. If production serves a broader user population, decision type, or data category than Annex VIII states, the compliance issue is not only monitoring quality. It is registration accuracy.

For agentic systems, agent identity, permission boundaries, least-privilege tool access, tool approval workflows, and tool-call audit trails help demonstrate that deployed capability remains within the registered intended purpose. Policy enforcement can block release or constrain runtime actions when registration or conformity status is missing or stale for a given version. Immutable audit logs of who approved registration data, what changed, and when updates were submitted support both internal assurance and external market-surveillance requests.

Runtime monitoring should detect drift: new tools, elevated permissions, purpose expansion, or environment promotion that was not reflected in the registration package. Detected drift should open a governance review tied to substantial-modification and registration-update workflows. Trussed AI focuses on runtime governance and security for enterprise AI agents, including agent identity, permissions, policy enforcement, tool governance, and audit logging. Those controls do not replace legal classification or Annex VIII filing, but they strengthen the evidence that declared purpose and operated behavior remain aligned.

Practical evaluation criteria

  • Classification defensibility. Can you show why each system is or is not in scope for Article 49, including Annex I exclusion logic where used?
  • Artifact traceability. Does every Annex VIII field resolve to controlled technical documentation, conformity outcomes, and a current declaration of conformity?
  • Release integrity. Can production deployment be blocked when registration or conformity status is invalid for that version?
  • Change coupling. Do model, purpose, and tool-capability changes automatically reopen registration accuracy review?
  • Operational match. Do runtime logs and permission boundaries evidence that live use matches the registered intended purpose?

Implementation checklist for AI governance leaders

  • Build an inventory of systems against Article 6(2)/Annex III and relevant Article 6(1) cases, with owner and version IDs.
  • Document provider versus deployer role, including any public-authority deployer registration duty.
  • Map Annex VIII fields to Annex IV technical documentation and the EU declaration of conformity.
  • Enforce a release gate: no market placement or putting into service without valid registration and conformity status.
  • Define substantial-modification and purpose-change triggers that force registration updates.
  • Retain evidence packs and use runtime identity, permissions, and tool-call logs to verify continued purpose alignment.

Article 49 registration FAQs

Do all high-risk AI systems require EU database registration?

Providers must register Annex III high-risk systems under Article 6(2) before market placement or putting into service. Annex I product systems register only in the limited Article 49 cases without third-party conformity assessment under that product law.

What is the relationship to the EU declaration of conformity?

Registration is linked to conformity assessment and the EU declaration of conformity. Annex VIII data should match the declared system identity, intended purpose, and conformity route documented under Articles 43, 47, and related technical documentation duties.

When must registration be updated after go-live?

Providers must keep registration information complete, correct, and up to date. Purpose changes, version changes, and substantial modifications should trigger re-validation and database updates, with renewed conformity work where required.

Are deployer organizations ever required to register?

Yes. Deployers that are public authorities, Union institutions, bodies, offices, or agencies, or acting on their behalf, must register use of specified high-risk AI systems under Article 49(3).

What remains uncertain operationally?

Commission database UI, API, and submission workflows may evolve in implementing guidance. Edge cases on substantial modification, component versus system registration, and mixed Annex I/Annex III products may need case-specific legal interpretation.

Align runtime operation with registered AI system purpose

Use runtime identity, permissions, policy enforcement, and audit logging to support ongoing Article 49 accuracy after initial EU database registration.

Explore Runtime Governance