Skip to content
Documentation · for startups

The EU AI Act compliance report, what it should contain.

There is no official certificate for most AI products. What buyers and investors want instead is a dated, written position. Here is what it contains and how to produce one.

Last updated 9 October 2026 · Regulation (EU) 2024/1689 as amended by the June 2026 Omnibus · Self-assessment, not legal advice

Sooner or later someone asks a startup for its "EU AI Act compliance report". It might be an enterprise buyer's procurement team, an investor's lawyer, or your own board. There is no official document by that name. The Act does not define a compliance report, and for most AI products there is no certificate to show. What the person asking wants is a written, dated position: what your AI does, which rules apply to it, and what you have done about them. This guide sets out what that document should contain, the ways to produce it, and how to keep it from going stale.

For most startups the honest headline of the report is short: "Limited or minimal risk, these three duties apply now, here is the evidence, and here is when we will look again." The value is in the reasoning and the date, not the length.

Why there is no certificate to send

Conformity assessment under Article 43 and CE marking exist for high-risk AI systems only. For an Annex III system they apply from 2 December 2027; for AI built into products covered by Annex I, from 2 August 2028. Limited-risk and minimal-risk systems, which is where most startup products land, have no conformity procedure and no certificate. So the document that answers the question is a classification and gap analysis you can stand behind, not a seal.

That is also why wording matters. A report that says "compliant" invites the reader to test the claim. A report that says "classified as limited risk on [date], Article 50(1) disclosure in place" gives the reader something to check.

What the report should contain

SectionWhat it statesBasis
System descriptionWhat the system does, for whom, and its intended purpose in one paragraphArt 3(1), 3(12)
ScopeWhy the Act applies, including where the output is usedArt 2
RoleProvider or deployer, per feature, and whether anything could shift that. Start from an AI inventoryArt 3(3), 3(4), 25
Prohibited practicesA check against each Article 5 practiceArt 5
GPAI checkWhether you are only using a general-purpose model or have become its providerArt 53
Risk tier with reasoningThe Annex III point engaged, or why none isArt 6, Annex III
Article 6(3) analysisIf an Annex III area is engaged: which condition you rely on, and a statement on profilingArt 6(3), 6(4)
Obligations and deadlinesEvery duty that applies, with its current post-Omnibus dateArts 4, 50, Chapter III
Gaps and prioritiesWhat is missing, in the order to fix itAll of the above
Date, version, limitsWhen it was done, against which text, and whether it is a self-assessment or a legal opinionRegulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744

The last row is the one most often missing. Deadlines moved in June 2026, so a report without a date and a version cannot tell the reader which rules it was written against. What the Omnibus changed.

The duties most startup reports end up listing

ObligationWhoApplies
Article 5 prohibited practicesEveryoneSince 2 Feb 2025
Article 4 AI literacyProviders and deployersSince 2 Feb 2025
Article 53 GPAI model dutiesModel providers onlySince 2 Aug 2025
Article 50(1) chatbot disclosureProvidersSince 2 Aug 2026
Article 50(2) synthetic content markingProviders of generative systems2 Dec 2026
Chapter III high-risk obligations, Annex IIIProviders and deployers of high-risk systems2 Dec 2027
High-risk obligations, Annex I productsProviders of regulated products2 Aug 2028

If your product has a chatbot, the Article 50(1) line is almost certainly in your report, and it is live. The free Article 50 disclosure generator drafts the wording, and the disclosure templates guide explains placement.

Three ways to produce it

RouteGood forWatch out for
Write it yourselfA founder who has read the Act and has a clear, low-risk productMissing the Article 6(3) profiling test or quoting pre-Omnibus dates; no structure a reviewer recognises
Structured self-assessmentGetting a consistent, dated document quickly, before counsel is neededIt reflects your answers. It is not legal advice, and should be described as a self-assessment
Lawyer's memoHigh-risk systems, borderline Annex III cases, contract disputes, regulator contactCost and turnaround; the lawyer will still need the same facts about your system

These are not exclusive. A common sequence is a structured self-assessment first, then a lawyer on the specific questions it flags. Counsel's time goes further when the classification groundwork is already written down.

How Reglynn's reports fit

Reglynn is one way to do the middle route. The free scan is eight questions plus an email step, about two minutes, and gives a likely risk tier on screen. The paid assessment is up to 33 questions, or 41 with the Document Pack, and produces a dated gap report covering the role analysis, the Article 6(3) and GPAI checks, the obligations that apply to your answers, each with its post-Omnibus deadline, a prioritised remediation list and GDPR overlap notes. Reports are generated and delivered automatically.

The sample report shows a complete one for a fictional HR-tech company, so you can judge the format before paying. It is a self-assessment: it is not a legal opinion, not a certification and not issued by a notified body.

Keeping it current

Start with the scope and tier. If the free scan or the high-risk checker shows nothing in Annex III, your report is mostly about Article 50, Article 4 and Article 5, and it can be short. If something is in Annex III, the report becomes your 2027 plan.

Frequently asked questions

Is there an official EU AI Act compliance certificate?

Only for high-risk systems. Conformity assessment under Article 43 and CE marking apply to high-risk AI systems, from 2 December 2027 for Annex III systems and 2 August 2028 for Annex I products. Limited-risk and minimal-risk systems have no certificate.

How long should an EU AI Act report for a startup be?

As long as the reasoning needs. For a limited-risk or minimal-risk product it can be a few pages: scope, role, tier with reasons, the Article 5 and GPAI checks, live duties with evidence, gaps and a date. A high-risk product needs a full workplan.

How often should the report be updated?

Whenever a feature changes its purpose or adds a prediction about people, and whenever the rules change. The Act was amended in June 2026, so a report dated before then may quote deadlines that no longer apply.

Start with your likely risk tier

The free scan takes about two minutes: eight questions plus your email. When you need the full document, the gap report sets out the obligations, Articles and post-Omnibus deadlines that apply to your answers, dated.

Start the free scan

No card, no account. Full gap report from €59. See a sample report. Prices exclude VAT, added at checkout where applicable. 14-day refund.

Self-assessment, not legal advice. This page and the Reglynn report are a structured self-assessment based on the EU AI Act (Regulation 2024/1689, as amended by the June 2026 Omnibus). They indicate where you likely stand and what to verify with a qualified advisor. Reglynn is not a notified body and issues no certification.

How this was made. Reglynn reports are generated by an AI system built on Anthropic's Claude and delivered automatically. Reglynn is a Limited Risk AI system under Article 50; we disclose AI involvement wherever it occurs, including in our chat assistant. Our own self-assessment.