Skip to content
Reference Guide AI Act & Italian Law Updated 20 Oct 2026

⚖️ EVIDE & AI Act

Evidence for Regulatory Requirements and Disputes

When the functioning of an AI system needs to be examined, it helps to be able to reconstruct the operational context, the available artefacts, the documented human oversight and the sequence of events. EVIDE contributes to preserving and making verifiable the material received, within a declared perimeter.

Section 1

1. Explore the detailed framework in your language

This page is EVIDE's English-language regulatory reference. The detailed evidentiary-defensibility framework it draws on is also available in four other languages, on EVIDE's AI Governance Dispute Defense Layer page.

Section 2 · Updated 20 October 2026

2. When the obligations apply

The AI Act applies in stages. Regulation (EU) 2026/1744 (the "AI Omnibus"), published in the Official Journal on 24 July 2026 and in force since 27 July 2026, postponed the high-risk rules and amended some provisions. The dates below follow the Regulation as amended.

FromWhat applies
2 February 2025Prohibited AI practices (Art. 5). AI literacy (Art. 4): as amended in 2026, providers and deployers take measures to support a sufficient level of AI literacy among the people dealing with AI systems on their behalf.
2 August 2025Obligations for general-purpose AI models, the governance framework and penalties.
2 August 2026General application of the Regulation, including the transparency obligations of Art. 50.
2 December 2026New prohibitions added in 2026 (AI-generated child sexual abuse material and non-consensual intimate content). Art. 50(2) marking obligations for generative systems already on the market before 2 August 2026.
2 December 2027High-risk rules for the stand-alone systems listed in Annex III (Art. 6(2)) - including risk management, logging, human oversight and deployer obligations.
2 August 2028High-risk rules for AI systems that are safety components of products covered by Section A of Annex I, or are themselves such products (Art. 6(1)).

Machinery after the 2026 amendments

The Machinery Regulation (EU) 2023/1230 has been moved from Section A to Section B of Annex I. For an AI system that is high-risk because it is a safety component of machinery, the AI Act's high-risk requirements therefore do not apply directly: only the provisions listed in Art. 2(2) apply, and AI-specific requirements are to be introduced through acts adopted under the Machinery Regulation. This does not take AI in machinery outside the AI Act as a whole: the rules that do not depend on Annex I - such as the prohibited practices, the transparency obligations of Art. 50 and, where the system's intended purpose falls within Annex III, the Annex III high-risk rules - continue to apply.

Systems already on the market (Art. 111)

  • High-risk systems placed on the market or put into service before the date from which the high-risk rules apply to them (2 December 2027 for Annex III systems, 2 August 2028 for Annex I systems) are subject to the Regulation only if, from that date, their design is significantly changed (Art. 111(2), as amended in 2026).
  • Exception: providers and deployers of high-risk systems intended to be used by public authorities must bring them into compliance by 2 August 2030, whether or not the design changes (Art. 111(2)).
  • Components of the large-scale EU IT systems listed in Annex X placed on the market or put into service before 2 August 2027 must comply by 31 December 2030 (Art. 111(1)).
  • General-purpose AI models placed on the market before 2 August 2025 must comply by 2 August 2027 (Art. 111(3)).
Most obligations in the tables below concern high-risk systems and therefore apply from 2 December 2027 or 2 August 2028, subject to the transitional rules above. Preserving records before those dates is not required by the Regulation; it can make later reconstruction easier once the obligations, or a dispute, arrive. National rules, such as the Italian legislation in Section 5, may already apply.

Section 3

3. Regulatory requirements and evidentiary questions

Regulatory obligations under the AI Act fall on specific addressees - mainly providers (who develop or have developed the AI system) and deployers (who use it under their own authority). The table below keeps that distinction explicit, and separates what the regulation requires from how EVIDE can help document it.

Regulatory areaEvidence questionHow EVIDE can help
Event logging & log preservation
Art. 12 (provider) · Art. 26(6) (deployer)
Providers must design high-risk systems to enable automatic event logging. Deployers must then keep those logs, to the extent under their control, for at least six months. Can the retained logs later be related to an independently anchored reference? EVIDE can receive and retain a submitted log file directly, computing its own hash and keeping it available for retrieval - or, for a log held externally, anchor the sender's declared reference and hash without itself verifying that the external file exists or matches it.
Human oversight documentation
Art. 14 (provider) · Art. 26(2) (deployer)
Providers must design for effective human oversight; deployers must assign it to competent, trained individuals with the necessary authority. For remote biometric identification, no action may be taken on a match unless it has been separately verified and confirmed by at least two natural persons (Art. 14(5), with limited exceptions). Can a specific instance of human review or intervention be shown, rather than only asserted? EVIDE can preserve a declared record of an oversight event - what was reviewed, by whom it was declared to be reviewed, what action followed - as submitted, within the agreed scope. Human Structured Intake structures that declaration: decision, declared role, declared closure time, readiness state and unresolved signals.
Risk management & technical documentation
Art. 9 · Art. 11 (provider)
Providers must maintain a risk management system and technical documentation covering the system's design, risks and mitigations. Can the state of that documentation at a given point in time be fixed and later verified? EVIDE can receive and retain submitted documentation directly, computing its own hash - or, for documentation held externally, anchor the sender's declared reference and hash without itself verifying the external material's existence or content.
Incident monitoring & reconstruction
Art. 72 · Art. 73 (provider) · Art. 26(5) (deployer)
Providers must monitor system performance after deployment and report serious incidents to the market surveillance authorities within deadlines that run from when they become aware of them (Art. 73); deployers must monitor operation and inform the provider or the authority of emerging risks and serious incidents. After an incident, can the sequence of declared events be reconstructed from material preserved independently of the system itself? EVIDE can preserve, as submitted within scope, the declared sequence of inputs, classifications, actions and human interventions associated with an event, so later reconstruction does not depend solely on the system's own account.
Explanation to affected persons
Art. 86 · Art. 26(11) (deployer)
The right to explanation is limited. It belongs to a person affected by a decision that the deployer takes on the basis of the output of a high-risk system listed in Annex III (except point 2, critical infrastructure), where the decision produces legal effects or similarly significantly affects that person in a way they consider to have an adverse impact on their health, safety or fundamental rights. The person can then obtain clear and meaningful explanations of the role of the AI system in the decision-making procedure and of the main elements of the decision taken. The right does not apply where Union law, or national law complying with it, provides exceptions or restrictions (Art. 86(2)), and applies only insofar as it is not already provided for under other Union law (Art. 86(3)). Separately, deployers of Annex III systems that make or assist decisions about natural persons must inform them that they are subject to the system (Art. 26(11)). In such cases, can what the system produced be told apart from what the person decided? A Human Structured Intake record preserves the human decision as declared at closure, with references to the AI output it considered, so the system's contribution and the human decision remain distinguishable later. EVIDE does not produce the explanation and does not judge whether it is adequate.
Fundamental rights impact assessment
Art. 27 (certain deployers)
Deployers that are public bodies or private entities providing public services, and deployers of systems for creditworthiness or life and health insurance (Annex III, points 5(b) and 5(c)), must assess the impact on fundamental rights before first use and notify the authority of the results. Can the version of the assessment in place at a given time be fixed? EVIDE can receive and retain the submitted assessment, computing its own hash, or anchor a declared reference and hash for an externally held copy. It does not assess the content or its adequacy.
External review & disputes
General - applies across providers and deployers
When a decision or outcome is challenged, can material be presented that was preserved before the dispute began, by a party independent of the one being examined? EVIDE's role is to preserve declared material independently and make its anchoring verifiable. What weight that material carries in a given review or dispute is for the examining authority, not EVIDE, to determine.
These rows describe candidate uses of EVIDE against documented AI Act obligations. Depending on how material is submitted, EVIDE either receives and retains the file itself or anchors a declared, externally-held reference; in neither case does preservation or anchoring by itself establish the truth of the content or that the declared events took place. These rows do not state that any particular organisation has adopted EVIDE for this purpose, and they do not substitute for legal advice on how a specific obligation applies to a specific system.

Section 4

4. High-risk areas and the human decisions they involve

Annex III lists the areas where stand-alone AI systems are classified as high-risk, subject to the conditions and exceptions of Art. 6(3). In each of them, the moment that is most often questioned later is a human decision taken with, or after, the system's output.

Annex III areaTypical human decision to document
1. BiometricsThe separate verification and confirmation of a remote biometric identification match before any action (Art. 14(5)).
2. Critical infrastructureContinuing, isolating or restoring operation when an AI safety component of digital infrastructure, road traffic or water, gas, heating or electricity supply raises an alert.
3. Education and vocational trainingConfirming or overriding an AI-assisted admission, assessment or exam-monitoring outcome.
4. Employment and workers managementConfirming or overriding an AI recommendation on recruitment, promotion, termination or task allocation.
5. Essential private and public servicesDeciding on eligibility for public benefits, on creditworthiness or on life and health insurance pricing, or confirming the priority of an emergency call. AI used to detect financial fraud is excluded from the creditworthiness point.
6. Law enforcementA decision by a law enforcement authority taken with the support of an AI system listed in point 6 - for example one used to assess the reliability of evidence - within the limits of the Regulation and of national law.
7. Migration, asylum and border control managementThe competent authority's decision on an application or a risk assessment, taken with the support of an AI system listed in point 7.
8. Administration of justice and democratic processesA decision by a judicial authority, or in alternative dispute resolution, prepared with an AI system that assists in researching and interpreting facts and the law. Point 8 also covers AI systems intended to influence the outcome of elections or referendums, under its own conditions.

From requirement to record. Human Structured Intake gives the person who closes such a decision a structured way to declare it: what was decided, in which declared role, at which declared closure time, under which threshold, whether an independent readiness gate was involved, and which signals remained unresolved. EVIDE preserves the declaration with its own intake timestamp and hash. It records what was declared; it does not establish that the oversight was effective or that the decision complied with the Regulation.

Human Structured Intake - use cases and fields ↗
This table summarises the Annex III areas for orientation only. Whether a specific system is high-risk depends on its intended purpose, on the exact wording of Annex III and on Art. 6(3); it requires a case-by-case assessment.

Section 5

5. Italian national legislation — D.Lgs. 160/2026

The obligations above come from Regulation (EU) 2024/1689 (the AI Act), which applies across the Union. Separately, Italy has adopted national legislation that is relevant to the same questions but operates under its own rules.

Italian national legislation — D.Lgs. 160/2026

Legislative Decree 9 September 2026, no. 160 (in force from 30 September 2026) introduces, among other things, Article 437-bis of the Italian Criminal Code and Articles 17 to 19 on civil proceedings. These are Italian national provisions. Whether and how they apply to a given system, organisation or dispute must be assessed case by case, with qualified legal advice.

Art. 437-bis of the Criminal Code: criminalises the omission of required technical security measures or human-oversight measures, and the alteration of a high-risk AI system, when concrete danger to life, to public or individual safety, or to State security results. Grave negligence and intentional omission by a professional user are separately addressed.

Article 17 of the decree (disclosure): on the application of the party claiming to have been harmed, who must present facts and elements making the claim plausible, the court can order the other party or a third party to produce the specifically relevant elements concerning how the AI system functioned - including the logs (AI Act Art. 12), the risk management documentation (Art. 9), the relevant parts of the technical documentation (Art. 11) and the information on human oversight (Art. 14). The order must be necessary and proportionate, and trade secrets are protected. An unjustified refusal by a party is assessed by the court as evidence under Article 116(2) of the Code of Civil Procedure; where it concerns the documentation listed above, the court, having assessed the other evidence, can treat the facts alleged by the claimant as admitted. A third party that fails to comply without justification can be fined from EUR 1,500 to 10,000.

Articles 18 and 19: where the damage results from a breach of one or more obligations under the AI Act, the causal link between the breach and the damage is presumed, unless proven otherwise; the claimant still has to show that the obligation applied, that it was breached and the damage (Art. 18). Conformity with the AI Act, even if certified, does not by itself exclude liability (Art. 19). Because several of the AI Act obligations referred to here apply to high-risk systems only from the dates in Section 2, their relevance depends on when the facts occurred.

Note: "Article 17" here refers to Article 17 of this Italian decree (access to evidence). It is a different provision from Article 17 of the EU AI Act itself, which concerns providers' quality management systems. The two should not be confused.

Section 6 · Standards update · 18 October 2026

6. AI standards and the evidence needed for later review

Quality management, organisational governance and AI security each raise practical questions about the records available for later examination. The standards below address those areas. The EVIDE examples describe potential uses of submitted records; they are not a clause-by-clause assessment against the standards.

EN 18286:2026 — Quality management for EU AI Act regulatory purposes

Commission Implementing Decision (EU) 2026/2274, published in the Official Journal on 14 October 2026, cites EN 18286:2026 - the first harmonised standard cited in support of the AI Act. It concerns the quality management system that providers of high-risk AI systems must put in place under Art. 17 of the AI Act, including post-market monitoring processes. Applying a harmonised standard is voluntary; conformity with it confers a presumption of conformity only for the requirements it covers, or the relevant parts of it - not for the AI system as a whole. In practice it becomes most relevant as the high-risk obligations start to apply (Section 2).

Potential EVIDE use: preserve submitted versions of procedures, declared review records, incident reports and corrective-action artefacts, together with their declared relationships, to support later examination of the documented sequence.

IEEE 2863-2026 — Organisational AI governance

Published on 16 October 2026, this Recommended Practice addresses governance activities for organisations developing or using AI. Its public description covers governance principles, processes for implementing them and case studies, including tensions between principles.

Potential EVIDE use: preserve a submitted account of a governance decision, the decision-maker's declared role, the criteria referenced and the supporting artefacts received. Preservation does not independently validate the person's authority or the merits of the decision.

ISO/IEC 27090:2026 — AI security threats and compromises

This international standard provides guidance on understanding, detecting and mitigating AI-specific security threats throughout the system life cycle, including threats such as data poisoning and model theft.

Potential EVIDE use: preserve submitted incident records, configuration snapshots, change records and response artefacts so their contents and declared context remain available for later examination.

After an event, which records allow a reviewer to reconstruct the applicable procedure, the declared decisions and the documented interventions?

Example: an organisation submits a report about an AI system, a record declaring that a human review took place, a revised procedure and related corrective-action artefacts. EVIDE can preserve the received files and compute their hashes, or anchor declared references and hashes for externally held material. Later review can distinguish what was declared, what was actually received and what remains unavailable.

These examples are based on the official public descriptions and the EU citation Decision. They do not establish EVIDE's conformity, certification or endorsement under any of these standards. A detailed mapping requires examination of the full standards and their applicable clauses. Neither the standards cited here nor the Decision establish a requirement to use EVIDE or an external evidentiary deposit.

Section 7

7. What EVIDE does not do

  • EVIDE supports reconstruction within the boundary of the material actually preserved.
  • EVIDE does not replace security measures, human oversight or risk management.
  • EVIDE does not automatically certify compliance with the AI Act.
  • EVIDE does not determine lawfulness, fault or responsibility.
  • EVIDE does not guarantee the admissibility or sufficiency of its records as evidence.
  • Anchoring alone does not automatically establish the truth of the original content.
External evidentiary deposit is not a general obligation under the AI Act. The Regulation does not require it, and this page does not present it as such. At the same time, internally produced logs are not without evidentiary value - they can be relevant. The distinction EVIDE addresses is about independent anchoring and later verification, not about whether internal records count at all.

References