What Is Annex IV Documentation? The Complete Guide

If your AI system is classified as high-risk under the EU AI Act, you must prepare Annex IV technical documentation before placing it on the EU market. This document is the single most important piece of evidence you will need if a regulator asks you to prove compliance.

Here is exactly what it requires.

What Is Annex IV?

Annex IV is an appendix to the EU AI Act (Regulation 2024/1689) that defines the minimum contents of the technical documentation high-risk AI providers must maintain.

The documentation must be prepared before* the system is placed on the market, *kept up to date throughout the system's lifecycle, and made available to national competent authorities on request.

The 8 Required Sections

1. General Description of the AI System

The system's intended purpose, including the specific context and conditions of use

The version number or other identifier

How the system interacts with hardware, software, and other AI systems

2. Description of the AI System Elements and Development Process

Computational resources used in development and deployment

Training methodologies — supervised, unsupervised, reinforcement learning, etc.

Architecture — model type (transformer, neural network, decision tree, etc.), number of parameters, key design choices

Data requirements — what input data the system requires and in what format

3. Training Data Information

Data sources and acquisition methods

Data governance practices — how data was labelled, cleaned, and validated

Data characteristics — demographic coverage, known limitations, potential biases

Any data augmentation techniques used

4. Validation and Testing Results

Metrics used to evaluate performance

Test results against those metrics, broken down by relevant subgroups

Known or foreseeable performance limitations

Steps taken to address bias or fairness issues

5. Monitoring, Functioning, and Control

What monitoring mechanisms are in place

How human oversight is implemented (Article 14)

Procedures for handling incidents or unexpected outputs

How the system can be halted or overridden

6. Risk Assessment

Identification of foreseeable risks to health, safety, and fundamental rights

Risk mitigation measures taken

Residual risks and why they are acceptable

7. Changes Made to the System

Log of significant changes made over the system's lifetime

How changes were assessed for impact on compliance

8. EU Declaration of Conformity

Reference to the signed declaration that the system meets EU AI Act requirements

Applicable standards used in the conformity assessment

Common Mistakes

Writing it after deployment — Annex IV documentation must be prepared before market placement, not after a regulator asks for it.

Not updating it — The documentation must reflect the current state of the system. A model update that changes accuracy by more than a few percent is a significant change that requires documentation updates.

Being too vague — "We use a machine learning model" is not acceptable. Regulators expect specifics: architecture type, training data size, demographic coverage of test data, specific performance metrics.

Missing subgroup analysis — If your AI makes decisions affecting people, regulators expect to see performance broken down by gender, age group, and ethnicity — not just overall accuracy.

How to Generate It Faster

Writing Annex IV documentation from scratch takes 2-4 weeks for a typical ML team. Guardia AI's Annex IV Generator reduces this to about 15 minutes by:

Asking the right questions and generating the document structure automatically

Pulling model architecture information from your code

Generating the required sections in the correct format for regulators

Generate your Annex IV documentation →