There is no single "AI Act certificate": for most systems, you prove compliance by producing, when an authority asks, the documentation and records required by your role and by the system's risk level. In practice that means an inventory of your AI systems, the documentation the Regulation requires, the logs generated by high-risk systems and a dated record of the checks and measures applied. What exactly you need depends on whether you are the provider or the deployer and on how each system is classified.
What can authorities ask for?
Market surveillance authorities supervise AI systems under Regulation (EU) 2019/1020, which Article 74 applies to the AI Act. For high-risk systems:
- providers must give them access to the documentation and, where relevant, the training, validation and testing data sets, including through APIs or other remote access where appropriate (Article 74(12));
- access to source code requires a reasoned request and is only possible once testing and auditing based on the provider's data and documentation have been exhausted or proved insufficient (Article 74(13));
- on a reasoned request, providers must supply the information and documentation showing conformity, in an official EU language the authority can easily understand, and give access to the logs under their control (Article 21).
Article 77, as amended by Regulation (EU) 2026/1744, gives authorities that protect fundamental rights (non-discrimination in particular) the power to request and access information and documentation created under the Regulation, in accessible language and machine-readable format, by electronic means. They go through the market surveillance authority, which can in turn request the material from the provider or the deployer.
Supplying incorrect, incomplete or misleading information can lead to fines of up to EUR 7.5 million or 1% of worldwide annual turnover (Article 99(5)). Documentation kept as you go, rather than rebuilt after the fact, lowers that risk.
Which documents and logs must you keep?
The main documentation duties concern high-risk systems and will apply from 2 December 2027 for Annex III (see the EU AI Act timeline).
| Item | Who | Legal basis |
|---|---|---|
| Technical documentation | Provider | Article 11 and Annex IV |
| Quality management system | Provider | Article 17 (harmonised standard EN 18286:2026) |
| Keeping documentation for 10 years | Provider | Article 18 |
| Automatic event logging | Built in by the provider | Article 12 |
| Keeping logs for at least 6 months | Provider and deployer, for logs under their control | Articles 19 and 26(6) |
| Documented "not high-risk" assessment of an Annex III system | Provider | Article 6(4) |
| Registration in the EU database | Provider; public-sector deployers | Article 49 |
| Fundamental rights impact assessment | Certain deployers | Article 27 |
Article 12 requires high-risk systems to technically allow automatic recording of events over their lifetime, so that risk situations can be identified, post-market monitoring carried out and operation monitored. Article 19 and Article 26(6) set a retention period suited to the system's purpose, of at least six months, unless EU or national law, in particular data protection law, provides otherwise.
The Omnibus keeps the registration of Annex III systems assessed as not high-risk under Article 6(3), but slims down the required content (recital 22).
Who must carry out a fundamental rights impact assessment?
Article 27 requires a fundamental rights impact assessment (FRIA) before an Annex III high-risk system is first deployed by bodies governed by public law, private entities providing public services, and deployers of creditworthiness or life and health insurance pricing systems (Annex III, points 5(b) and (c)). Critical infrastructure (point 2) is excluded.
The assessment describes the processes involved, the period and frequency of use, the categories of people affected, the specific risks, the human oversight measures and what will be done if risks materialise. The results are notified to the market surveillance authority. Since the Omnibus, deployers can cross-reference the relevant parts of an existing data protection impact assessment (DPIA), and the AI Office is to provide a questionnaire template.
What about systems that are not high-risk?
No technical file is required, but you still need to be able to show that the applicable duties are met: the Article 50 disclosure (screenshot of the opening screen, the text shown, the date it went live), the marking of generated content, the Article 4 AI literacy measures (training, materials, attendees) and the absence of any Article 5 prohibited practice. For chatbots, see is my chatbot covered?.
What a register and an evidence trail are, and are not
A register (inventory) of AI systems lists each system, its purpose, your role, its classification and the resulting obligations. The AI Act does not require one as such from most businesses, and it is not the same thing as the EU database under Article 49. It is still the indispensable starting point: without an inventory, you cannot know which obligations apply.
An evidence trail records, with timestamps, classification decisions, the controls applied and their results. If each entry is chained to the previous one with a cryptographic hash, any later alteration becomes detectable.
Neither is a certification. They do not replace the Annex IV technical documentation, the quality management system or the logs the system itself must generate under Article 12. An evidence trail proves that a check was carried out and when; on its own, it does not prove that the classification was right or that the measure was sufficient.
VerifactAI keeps that register and trail: each decision is written to a SHA-256 hash-chained log, and the tool identifies the obligations, records the controls applied and the evidence, and exports a printable and machine-readable dossier. Prompts and model responses never pass through its API.
Sources
- Regulation (EU) 2024/1689, EUR-Lex
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex
- Article 74, AI Act Service Desk
- Article 77, AI Act Service Desk
- Article 27, AI Act Service Desk
- Implementing Decision (EU) 2026/2274 on harmonised standards, EUR-Lex
This article is general information, not legal advice.