EU AI Act and AI flows
Regulation · EU AI Act
EU AI Act: assess roles, applicable dates and evidence for an AI mission
An integrator is preparing a client mission: which uses are permitted, who decides and what must remain available if the work changes? The AI Act introduces a prior question: which system is involved, what is its intended purpose and what role does each organisation perform? Calling something an “agent” does not answer that question.
This article provides preparation guidance for the European Union framework, reviewed on 20 October 2026. It separates the regulation’s provisions, matters for competent assessment and DUBSAR’s intended role.
In brief
The question and the useful result.
- Question
- Which AI Act provisions concern this system, in which role and from which date?
- Useful result
- An assessment and timeline record for competent review, separate from operational documentation or a compliance verdict.
EU AI Act and AI flows
Use the timetable currently adopted
Read the text consolidated on 27 July 2026, particularly Article 113, with its exceptions and transitions. The general application date is 2 August 2026. Chapter III Sections 1–3 on high-risk systems apply, subject to the text’s qualifications, from 2 December 2027 for Article 6(2) and Annex III, and 2 August 2028 for Article 6(1) and Annex I.
The Commission confirms that the AI Omnibus entered into force on 27 July 2026. Those extensions should no longer be described merely as a proposal. They do not postpone every other provision. Article 111 also contains transitions, including one until 2 December 2026 for Article 50(2) concerning certain systems placed on the market before 2 August 2026.
EU AI Act and AI flows
Assess the system before assigning obligations
Articles 3 and 6 and the annexes help assess provider and deployer roles and system classification. An integrator is not automatically only a technical contractor: changes and the conditions of making the system available need examination. Nor should obligations on general-purpose AI models be attributed to every user of those models.
Within the high-risk framework, Articles 12, 14 and 26 address recorded events, human oversight and deployer obligations. They do not impose a uniform duty to retain all prompts indefinitely. Logging scope, control and retention require assessment against the case and data-protection rules.
Article 50 concerns transparency for certain systems and uses: informing people, marking or disclosure according to the actor’s role and the content. An internal mission log does not establish that required information or marking actually reaches its intended recipient.
EU AI Act and AI flows
A record for competent assessment
| Field | Decision and reference to prepare |
|---|---|
| System and purpose | Version, intended outcome, affected people and excluded uses. |
| Roles | Provider, deployer and other actors; assessment owner and supporting information. |
| Classification | Examined articles and annexes, reasoning, open questions and changes requiring reassessment. |
| Timetable | Specific provision, applicable date and any transition; source and review date. |
| Implementation | Control to establish, authorised decision-maker and observable evidence of operation. |
| Handover | Mission state, accessible documents, limits and the next assigned check. |
This is a preparation method usable without DUBSAR. It replaces neither legal assessment nor documentation or evaluation required for the system concerned.
EU AI Act and AI flows
Scenario: an administrative workflow changes purpose
In this illustrative example, a mission prepared interview invitations after selection had already been decided. The client now proposes using it to rank applicants. The integrator retains that change as a decision requiring review rather than absorbing it into a prompt edit.
Appropriate specialists reassess purpose, actor roles, classification and applicable dates. Authorisation for the previous mission does not automatically cover the new one. The example does not itself settle the system’s legal classification.
For continuation, retain the assessed version, decisions, restrictions and outstanding evidence. DUBSAR aims to connect those elements. It does not establish automatic classification, regulatory marking or demonstrated compliance.
EU AI Act and AI flows
Frequently asked questions
Are all AI agents high-risk systems?
No. Assessment depends partly on intended purpose, actor role and the regulation’s categories. An agent architecture or model name is insufficient. Establish the conclusion for the specific system and use.
Does the high-risk extension suspend every obligation?
No. The consolidated timetable distinguishes provisions and transitions. Identify the one relevant to the case; other applicable rules, including personal-data requirements, do not disappear.
Does a record or DUBSAR approval establish AI Act compliance?
No. A record prepares decisions and checks. Responsible actors must assess and implement the requirements. This page establishes no automatic classification, conformity assessment or DUBSAR certification.
EU AI Act and AI flows
Where DUBSAR fits
DUBSAR is developing a governance foundation for integrators and teams deploying agents for clients: durable mission state, explicit authorisations, human decisions, evidence and continuation of work. It aims to connect what was requested, what was permitted and what is established about the outcome.
Selected components and workflows have been tested; full B2B integration and validation are still underway. Pilots are not currently open.
EU AI Act and AI flows
Sources and scope of review
Official sources reviewed on 20 October 2026. Scenarios and records are preparation methods that can be used without DUBSAR. They establish neither an available integration nor compliance.
- EUR-Lex — Regulation (EU) 2024/1689, consolidated 27 July 2026 — Current consolidated AI Act, including amended dates and transitional provisions.
- Commission européenne / European Commission — AI Omnibus enters into force — Entry into force on 27 July 2026; high-risk dates moved to December 2027 and August 2028.
EU AI Act and AI flows