International Standards for Conformity Assessment, Certification and Accreditation
The report examines international standards for conformity assessment, certification and accreditation in the field of AI Assurance. It asks what evidence can justifiably confirm a specific property of an artificial intelligence system and what legal meaning such confirmation carries. AI Assurance is treated as a set of procedures with different objects: a management process, a technical property of a system, a specific claim or the competence of an assessor.
The normative part distinguishes audit, testing, verification, certification, accreditation and notification. It analyses ISO/IEC 42001, 42006, 17021-1, 17065, 17025, 17029 and 17011; the transition of the global mutual recognition system to Global Accreditation Cooperation Incorporated from 1 January 2026; the European accreditation framework under Regulation (EC) No 765/2008; and the 2026 version of the EU AI Act, including the assessment routes under Annexes VI and VII and the role of notified bodies. The approaches of the United Kingdom, the United States, Singapore and the OECD are examined for comparison. The methodology combines metrology, statistics, quality management theory and comparative legal analysis.
The main conclusion is that the adequacy of assurance depends on the substantive link between criterion, method and decision. A management system certificate does not automatically confirm the quality of every output, and a successful test of a single function does not establish the lawfulness of the entire use. Mutual recognition of assessors’ competence operates within its defined scope and does not displace the obligations of the receiving jurisdiction. These points are illustrated by a hypothetical dossier for a legal assistant and by scenarios for banks and sanctions screening.
For ARGA, the report proposes a research programme for reviewing the sufficiency of evidence without claiming the status of an accreditation body or a notified body: a claim passport, a rule for handling contradictions and a pilot test of the reproducibility of expert assessments. It outlines three possible analytical products (Regulatory Route Memorandum, Evidence Sufficiency Review and Third-Party Assurance Due Diligence) and transparency proposals for international discussion, including a scope-of-assessment passport and the disclosure of statistical uncertainty.
The report relies on publicly available sources as of 21 October 2026; the full paid texts of the standards were not used. Its models, numerical examples and forms are methodological tools developed for the study and do not indicate that ARGA holds any accreditation or has actually tested real systems.
Citation Rules
- Mandatory source attributionWhen using ARGA Observatory materials, the full name must be cited.
- Date and version indicationFor analytical reports, the year of publication is mandatory.
- Link to the originalElectronic materials must be accompanied by an active link to the official website.
- Context preservationCitations must not be shortened or altered in a way that distorts the original meaning.
- Note on adaptationIf the text is abridged or translated, state: "adapted from ARGA Observatory report".
- No commercial use without written permission from the organization
- Data accuracy preservationCharts and tables must be reproduced without changes.
Ask the ARGA experts
Four short steps. We reply within one business day.