Article 11 of the EU AI Act requires every provider of a high-risk AI system to draw up technical documentation before the system is placed on the market, and Annex IV lists the nine sections that file must contain at a minimum. The technical documentation is the evidence base for conformity assessment, the document a market surveillance authority will ask for first, and the record that must be kept for ten years. This guide walks through Article 11, each Annex IV point, how the file is used in internal control and notified body assessment, and how to build it so that it stays current. It does not repeat the assessment routes and harmonised standards covered in the EU AI Act Conformity Assessment and Harmonised Standards Guide, or the substantive requirements covered by the Article 9 to 15 guides linked below.
This page is general information, not legal advice. Status as of 17 Nov 2026, based on Regulation (EU) 2024/1689 as published in the Official Journal. Check EUR-Lex for any later amending acts, delegated acts under Article 11(3) and Commission guidance before relying on dates or content.
Key takeaways
- Article 11(1) requires the technical documentation to be drawn up before the high-risk system is placed on the market or put into service, and kept up to date.
- Annex IV sets nine minimum content points, from a general description through development, testing, risk management, standards, the declaration of conformity and post-market monitoring.
- SMEs, including start-ups, may provide the Annex IV elements in a simplified manner, using a form the Commission is to establish, which notified bodies must accept.
- Where the system is covered by Union harmonisation legislation in Annex I Section A, the provider draws up a single set of technical documentation.
- Article 18 requires the file to be kept for 10 years after placing on the market or putting into service.
- Under Article 113 most high-risk rules apply from 2 August 2026, and Article 6(1) product-related high-risk obligations from 2 August 2027.
- Breaches of provider obligations can attract fines of up to EUR 15 million or 3% of worldwide annual turnover, and misleading information to authorities up to EUR 7.5 million or 1%.
Article 11 and Annex IV at a glance
| Item | Provision | Summary |
|---|---|---|
| Duty to document | Article 11(1) | Draw up before placing on the market or putting into service; keep up to date; contain at least the Annex IV elements |
| SME simplification | Article 11(1) | Simplified provision of Annex IV elements using a Commission form |
| Single file | Article 11(2) | One set of documentation for products under Annex I Section A legislation |
| Amendment | Article 11(3) | Commission may amend Annex IV by delegated act |
| Provider duty | Article 16(d) | Keep the documentation referred to in Article 18 |
| Retention | Article 18(1) | 10 years after placing on the market or putting into service |
| Authority access | Article 21 | Provide documentation on reasoned request in a language easily understood by the authority |
| Assessment | Annex VI and Annex VII | Examined by the provider in internal control, or by a notified body |
| Application | Article 113 | 2 August 2026 general; 2 August 2027 for Article 6(1) |
Scope
The documentation duty sits with the provider of a high-risk AI system as classified under Article 6, whether through Annex III use cases or because the AI system is a safety component of a product covered by Annex I legislation. Deployers do not draw up Annex IV files, but they rely on the instructions for use that form part of it, and an organisation that substantially modifies a high-risk system or puts its name on one can take on provider obligations. The EU AI Act Provider vs Deployer Obligations Guide explains those role changes.
Financial institutions subject to Union financial services law may keep the technical documentation as part of the records they already maintain under that law, under Article 18(3). That changes where the file lives, not what it contains.
The file is per system and per version. Annex IV point 1(a) requires the version to be stated and its relation to previous versions to be shown, so the documentation must track the system's lifecycle rather than describe it once.
Key obligations: the nine Annex IV points
Annex IV requires the following, as applicable to the AI system. The headings below follow the Annex numbering.
1. General description
The intended purpose, the provider's name and the system version with its relation to earlier versions; how the system interacts with hardware or software outside it, including other AI systems; relevant software or firmware versions and update requirements; all forms in which it is placed on the market, such as embedded software, downloads or APIs; the hardware it runs on; for product components, photographs or illustrations of external features, markings and internal layout; a basic description of the user interface provided to the deployer; and the instructions for use. Point 1(b) is where integrations with other AI systems and agents, tools and external services belong, and it is often written too thinly.
2. Elements and development process
This is the core of the file. It covers the methods and steps used in development, including the use of pre-trained systems or tools from third parties and how they were used, integrated or modified; design specifications, including general logic and algorithms, key design choices with rationale and assumptions, including about the persons or groups the system is intended for, main classification choices, what the system is designed to optimise for, the expected output and output quality, and trade-offs made to meet the Chapter III Section 2 requirements; the system architecture showing how software components build on or feed into each other, and the computational resources used to develop, train, test and validate it; data requirements where relevant, including training methods, datasheets on provenance, scope and main characteristics, how data was obtained and selected, labelling and cleaning methods; the assessment of human oversight measures under Article 14, including technical measures that help deployers interpret outputs under Article 13(3)(d); pre-determined changes to the system and its performance, and the technical solutions that keep it compliant; validation and testing procedures, data, metrics for accuracy, robustness and other requirements, and potentially discriminatory impacts, with test logs and reports dated and signed by the responsible persons; and the cybersecurity measures put in place.
3. Monitoring, functioning and control
The capabilities and limitations in performance, including accuracy for specific persons or groups and the overall expected accuracy for the intended purpose; foreseeable unintended outcomes and sources of risk to health, safety, fundamental rights and discrimination; the human oversight measures under Article 14, including tools for interpreting outputs; and specifications on input data, as appropriate.
4. Appropriateness of performance metrics
Why the chosen metrics are appropriate for this specific system. A list of metrics is not enough; the file must justify them against the intended purpose.
5. Risk management system
A detailed description of the risk management system under Article 9. See the EU AI Act Article 9 Risk Management System Guide.
6. Lifecycle changes
A description of relevant changes made by the provider through the system's lifecycle.
7. Harmonised standards and other solutions
A list of harmonised standards applied in full or in part whose references are published in the Official Journal. Where none have been applied, a detailed description of the solutions adopted to meet the Chapter III Section 2 requirements, including other standards and technical specifications applied.
8. EU declaration of conformity
A copy of the declaration drawn up under Article 47.
9. Post-market monitoring
A detailed description of the system in place to evaluate performance in the post-market phase under Article 72, including the post-market monitoring plan referred to in Article 72(3). See the EU AI Act Post-Market Monitoring Guide (Article 72).
How the file is used in conformity assessment
Under internal control in Annex VI, the provider itself examines the information in the technical documentation to assess compliance with the essential requirements in Chapter III Section 2, and verifies that the design and development process and the post-market monitoring are consistent with it. The file is therefore both the subject and the record of the self-assessment.
Under Annex VII, the provider applies to a notified body for assessment of the technical documentation, and the application includes the Annex IV file. The notified body examines it and may be given access to training, validation and testing datasets, may ask for further evidence or tests, and in limited cases may access the trained model. Changes that could affect compliance must be assessed by the notified body that issued the certificate. Either way, gaps in the file translate directly into assessment findings.
Article 18 lists what must be kept alongside the technical documentation for 10 years: the quality management system documentation under Article 17, documentation of changes approved by notified bodies, decisions and documents issued by notified bodies, and the EU declaration of conformity. Member States set conditions for keeping it available if a provider goes bankrupt or stops trading before the end of that period.
Enforcement and penalties
Under Article 21 a provider must, on a reasoned request from a competent authority, provide the information and documentation needed to demonstrate conformity, in a language the authority can easily understand, and give access to automatically generated logs under its control. Information obtained is subject to the confidentiality rules in Article 78.
Article 99(4) allows fines of up to EUR 15 000 000 or 3% of total worldwide annual turnover, whichever is higher, for non-compliance with listed operator obligations, including the provider obligations in Article 16, which include keeping the documentation. Article 99(5) allows fines of up to EUR 7 500 000 or 1% for supplying incorrect, incomplete or misleading information to notified bodies or national competent authorities. For SMEs, including start-ups, Article 99(6) caps each fine at the lower of the percentage or amount. A file that overstates testing or omits known limitations is therefore a risk in itself, not just a weaker file.
What it means for security and identity teams
Point 2(h) cybersecurity is more than a paragraph
Annex IV point 2(h) asks for the cybersecurity measures put in place, and Article 15 sets the substantive requirement, including resilience against attempts to alter use, outputs or performance through data poisoning, model poisoning, adversarial examples or model evasion, confidentiality attacks and model flaws. The documentation should reference the threat model, the security controls on the training pipeline, model registry and inference endpoints, and the test evidence. See the EU AI Act Article 15 Guide.
Document the non-human identities around the system
Points 1(b), 2(a) and 2(c) together describe interactions, third-party components and architecture. For a modern AI system that means service accounts for training jobs, API keys for model providers, tokens used by agents and tools, and credentials for data sources. Recording which identities can change the model, its data or its configuration, and how they are controlled, gives the authority evidence that the system cannot be altered outside the documented process, and supports the lifecycle change records in point 6.
Integrity of the file itself
Point 2(g) requires test logs and reports to be dated and signed. Store the file and its evidence in a system with access control, versioning and tamper evidence, and sign artefacts with identities bound to named responsible persons. The automatic logs in Article 12 and the documentation in Annex IV should cross-reference each other; see the EU AI Act Article 12 Guide.
Supply chain evidence
Where the system uses a third-party pre-trained model or general-purpose AI model, point 2(a) requires a description of how it was used, integrated or modified. Collect the upstream provider's documentation, model version and hashes, and record the evaluation performed after integration.
Compliance steps
- Confirm classification and role. Record why the system is high-risk under Article 6 and that your organisation is its provider.
- Set up the file structure. Create one section per Annex IV point, with an owner for each and a version tied to the system release.
- Draft the general description. Include every distribution form, integration and the instructions for use before design work ends.
- Capture development evidence as you build. Design decisions, data sheets, architecture diagrams and compute resources are cheaper to record at the time than to reconstruct.
- Run and sign tests. Produce dated, signed test reports for accuracy, robustness, cybersecurity and discriminatory impact, including for pre-determined changes.
- Link the risk, oversight and post-market sections. Reference the Article 9 risk file, Article 14 measures and the Article 72 plan rather than duplicating them inconsistently.
- List standards. Record harmonised standards applied, or describe the alternative solutions in detail.
- Attach the declaration of conformity. Add the Article 47 declaration once the assessment is complete.
- Plan for change and retention. Define what triggers an update, how notified body approvals are recorded, and how the file will be kept for 10 years.
Common failure patterns
- The file is written after release from memory, so design rationale and data provenance are vague.
- Performance metrics are listed without the point 4 justification of why they suit the purpose.
- Accuracy is reported only in aggregate, with no figures for specific persons or groups.
- Third-party models and tools are named but their integration and modification are not described.
- Cybersecurity is a generic paragraph with no threat model or test evidence.
- Test reports are not dated or signed.
- The file is not versioned, so it describes a model that is no longer in production.
- No plan exists for keeping the file for 10 years if the product line or company ends.
Practitioner checklist
- Map every Annex IV point and sub-item to a named section and owner.
- Tie each file version to a system version, and show its relation to previous versions.
- Record every third-party pre-trained model or tool and how it was modified.
- Keep data sheets covering provenance, scope, selection, labelling and cleaning.
- Hold dated, signed test reports, including discriminatory impact testing.
- Document the security controls and the non-human identities that can change the model, data or configuration.
- Reference the Article 9 risk management system and the Article 72 monitoring plan.
- List harmonised standards applied, or describe the alternative solutions.
- Store the file with versioning, access control and a 10-year retention plan.
- Confirm the file can be supplied promptly in the language an authority requests.
Standards and references
- EUR-Lex: Regulation (EU) 2024/1689 (Artificial Intelligence Act), Articles 11, 16, 18, 21, 99, 113 and Annexes IV, VI and VII
- ISO: ISO/IEC 42001:2023 Artificial intelligence management system
- NIST: AI 100-1 Artificial Intelligence Risk Management Framework (AI RMF 1.0)
- EU AI Act Conformity Assessment and Harmonised Standards Guide
- EU AI Act High-Risk AI Systems Guide
- EU AI Act Article 10 Data and Data Governance Guide
- ISO/IEC 42001 Annex A Controls Guide
Related NHI Mgmt Group resources: EU AI Act Conformity Assessment and Harmonised Standards Guide · EU AI Act High-Risk AI Systems Guide · EU AI Act Article 9 Risk Management System Guide · EU AI Act Post-Market Monitoring Guide (Article 72)