Annex VI Internal Controls
under the EU AI Act

Annex VI of the EU AI Act sets out the "internal control" procedure - a self-assessment route that allows providers of most high-risk AI systems to demonstrate conformity without mandatory third-party (notified body) involvement. It is the default conformity assessment pathway for the majority of Annex III high-risk systems and represents the least burdensome route when applicable. This article articulates when and how the Annex VI internal controls route applies, its detailed requirements, practical application, advantages and limitations, edge cases, and the interplay with prEN 18286 once cited.

Regulatory Foundation: When Annex VI Internal Controls Applies

Article 43(1) and Annex VI establish that the internal control procedure is the default conformity assessment route for high-risk AI systems listed in Annex III, except where:

• the system falls under Annex III point 1 (biometric identification/verification in certain sensitive contexts) or is a safety component of a product listed in Annex I subject to third-party conformity assessment under Union harmonisation legislation - these require notified-body involvement (Annex VII);
• harmonised standards or common specifications are not applied in full, and the provider voluntarily chooses or is required to use the Annex VII route;
• the provider opts for notified-body assessment even when not mandatory (e.g., to increase credibility or reduce market-surveillance risk).

In practice, Annex VI applies to the large majority of high-risk systems in Annex III points 2-8 (critical infrastructure, education, employment, essential services, law enforcement, migration, justice), provided harmonised standards (e.g., future prEN 18286) or common specifications are applied where available, or the provider can demonstrate full compliance through internal evidence alone.

Requirements of the Annex VI Internal Controls Procedure

Annex VI requires the provider to perform a complete internal conformity assessment before placing the high-risk AI system on the market or putting it into service. The core obligations include:

1. Quality Management System (QMS) Compliance: Demonstrate that a QMS meeting Article 17 (including the 13 elements) is established, implemented, documented, and maintained.
2. Technical Documentation: Prepare and maintain comprehensive technical documentation per Annex IV, covering system description, development process, risk management details, data governance, validation results, robustness measures, and post-market monitoring plan.
3. Internal Verification: Verify through internal review that the system complies with all essential requirements (Chapter III, Section 2) and that the QMS ensures ongoing conformity.
4. EU Declaration of Conformity: Draw up and sign a formal declaration attesting full compliance, including unique identifier, standards applied (or justification for non-application), and responsible person details.
5. CE Marking & Registration: Affix the CE mark and register the system in the EU database before deployment.
6. Ongoing Obligations: Maintain documentation for 10 years (Article 18), operate post-market monitoring (Article 72), and report serious incidents (Article 73).

When prEN 18286 is cited in the Official Journal, full alignment with it grants presumption of conformity for the QMS element - dramatically reducing the evidential burden during internal verification and future market surveillance audits.

Practical Application: How Providers Execute Annex VI Internal Controls

The internal controls route places full responsibility on the provider but offers flexibility and cost/time savings when executed well:

• Documentation & Evidence: Maintain a single, traceable compliance file (often digital) linking quality policy ? risk assessments ? validation records ? post-market plan.
• Internal Review Process: Conduct structured internal audits/checklists covering all essential requirements; involve cross-functional team (engineering, quality, legal, risk).

Examples:

• An employment screening AI (Annex III point 4) uses Annex VI: builds QMS aligned with prEN 18286 draft, prepares Annex IV technical file showing bias mitigation and validation, performs internal verification, issues declaration, and registers - all without notified-body involvement.
• A critical-infrastructure predictive maintenance AI (point 2) follows the same route: internal controls suffice as long as harmonised standards are applied or full compliance is demonstrable.

Advantage: Faster to market, lower cost (€10k-€50k vs. €50k-€200k+ for notified-body), full control over timing. Limitation: Higher burden of proof during market surveillance - authorities can challenge any element at any time.

Edge Cases and Nuances When Annex VI Applies (or Does Not)

Applies:

• Most Annex III points 2-8 systems when standards/common specifications are followed or full internal evidence is robust.
• Systems where the provider already has a mature QMS (e.g., ISO 9001 certified) and can map it to Article 17 requirements.

Does Not Apply / Not Recommended:

• Biometric systems in sensitive contexts (Annex III point 1) - Annex VII mandatory.
• Safety components of Annex I products (medical devices, machinery) requiring third-party assessment under sectoral law.
• Cases where the provider seeks maximum external credibility (e.g., high-visibility applications, insurance purposes) or wants to minimise market-surveillance challenge risk.

Edge case: Continuously learning systems using Annex VI must document robust change-control and post-market monitoring procedures internally; auditors/authorities will scrutinise these heavily due to dynamic risk profile.

Integration, Strategic Implications, and Link to prEN 18286

Annex VI integrates directly with prEN 18286: once cited, alignment with the standard provides presumption of conformity for the QMS element, making internal verification substantially easier and more defensible. Providers should map existing processes to prEN 18286 draft Clauses 5-10, conduct gap analyses, and prepare internal audit trails that mirror notified-body expectations - even when using the self-assessment route.

Strategic implications: Choosing Annex VI accelerates time-to-market and reduces cost for eligible systems, but requires strong internal governance and documentation discipline. Missteps can lead to post-market enforcement actions. Best practice: Document classification rationale clearly, implement a risk-based internal audit programme, use prEN 18286 draft as a checklist for readiness, and maintain a living compliance file ready for unannounced authority requests.

In summary, Annex VI internal controls is the proportionate, self-assessment route that applies to most high-risk AI systems in Annex III points 2-8. It demands rigorous internal verification of QMS and essential requirements compliance, achieves faster and less costly market access when executed well, and - once prEN 18286 is cited - benefits from strong presumption of conformity, balancing innovation with robust regulatory oversight.

Content based on the EU AI Act (Regulation (EU) 2024/1689), Articles 17 & 43, Annexes VI & VII, and the prEN 18286 draft standard. The standard remains under revision; always consult the latest CEN/CENELEC drafts, EU AI Office guidance, and legal experts for implementation. High-risk provisions become fully applicable on 2 August 2026.