{"id":102,"date":"2026-06-23T10:00:00","date_gmt":"2026-06-23T08:00:00","guid":{"rendered":"https:\/\/garion-ai.de\/blog\/?p=102"},"modified":"2026-06-23T10:00:00","modified_gmt":"2026-06-23T08:00:00","slug":"eu-ai-act-annex-iv-technical-documentation","status":"publish","type":"post","link":"https:\/\/garion-ai.de\/blog\/eu-ai-act-annex-iv-technical-documentation\/","title":{"rendered":"EU AI Act \u2013 Annex IV: What belongs in your technical documentation"},"content":{"rendered":"<p>Anyone placing an AI system on the EU market under the AI Act will need technical documentation. For high-risk systems, <strong>Annex IV<\/strong> defines exactly what that documentation must contain. The good news: at heart, the annex is a structured checklist. The bad news: much of it cannot be produced on demand at the end \u2014 it has to be carried along across the whole lifecycle.<\/p>\n<h2>What Annex IV requires<\/h2>\n<p>Annex IV lists the components of the technical documentation. Simplified, they fall into six blocks:<\/p>\n<ul>\n<li><strong>General description<\/strong> of the system: intended purpose, responsible parties, versions, interaction with hardware and software.<\/li>\n<li><strong>Detailed technical description<\/strong>: development steps, system architecture, compute resources, data requirements and the design choices made.<\/li>\n<li><strong>Data governance<\/strong>: origin, preparation, assumptions and known biases of the training, validation and test data.<\/li>\n<li><strong>Monitoring, functioning and control<\/strong>: accuracy, robustness, metrics and the limits of the system.<\/li>\n<li><strong>Risk management<\/strong>: the system established under Article 9 and the measures taken.<\/li>\n<li><strong>Changes and conformity<\/strong>: lifecycle changes, standards applied and the EU declaration of conformity.<\/li>\n<\/ul>\n<h2>Why readiness starts early<\/h2>\n<p>The classic mistake is to treat documentation as a closing act. In practice, data provenance, design rationale or test results can rarely be reconstructed cleanly after the fact. Teams that instead adopt the Annex IV structure early \u2014 as their working outline \u2014 capture evidence where it is created: inside the development process itself.<\/p>\n<p>In concrete terms, that means a living document rather than a point-in-time report. Every relevant decision \u2014 an architecture choice, a rejected dataset, a re-tuned threshold \u2014 is recorded together with its rationale. It is exactly this chain of reasoning that later makes a conformity assessment traceable.<\/p>\n<h2>A pragmatic sequence<\/h2>\n<p>For teams without a dedicated compliance office, a lean order of work tends to hold up:<\/p>\n<ul>\n<li>First, pin down the <strong>intended purpose<\/strong> precisely \u2014 it drives almost everything else.<\/li>\n<li>Then document the <strong>data basis<\/strong> while the sources are still fresh and at hand.<\/li>\n<li>In parallel, set up <strong>risk management<\/strong> and let it run alongside development.<\/li>\n<li>Only at the end, bring together the cross-cutting artefacts \u2014 metrics, declaration of conformity, references to standards.<\/li>\n<\/ul>\n<h2>Where GARION fits in<\/h2>\n<p>GARION maps the Annex IV structure as a readiness domain: it walks you through the required blocks, records decisions with their rationale, and shows the working status per section. The output is not a conformity verdict but an ordered, auditable state of preparation \u2014 the basis on which professionals carry out the actual assessment.<\/p>\n<hr>\n<p><em>Note: GARION supports <strong>preparation<\/strong> for the EU AI Act and related frameworks. GARION is not a substitute for legal advice, conformity assessment or certification.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Annex IV is, at heart, a structured checklist for high-risk AI technical documentation \u2014 best kept as a living document from day one.<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"garion_pair":"annex-iv","garion_lang":"en","footnotes":""},"categories":[2],"tags":[],"class_list":["post-102","post","type-post","status-publish","format-standard","hentry","category-eu-ai-act"],"_links":{"self":[{"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/posts\/102","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/comments?post=102"}],"version-history":[{"count":0,"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/posts\/102\/revisions"}],"wp:attachment":[{"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/media?parent=102"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/categories?post=102"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/garion-ai.de\/blog\/wp-json\/wp\/v2\/tags?post=102"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}