Sources accessed on 29 October 2026 are listed below. The explanations summarise public information; readers should consult the linked sources for their full scope and current status.
Why this page exists
Claims that AI is "certified", "audited", "compliant" or "aligned" can describe different things: an organisation's management system, a particular system or an organisation's own assertion. This page explains eight examples and their limits. The explanations assume that any underlying certificate or report is genuine and that its stated scope and status have been confirmed. A claim alone does not establish those facts. This is general information, not legal or insurance advice.
1. "We are certified to ISO/IEC 42001"
What it establishes. A certification body has audited the organisation's AI management system and decided to certify it. ISO/IEC 42001:2023 specifies requirements for how an organisation manages its AI activities and improves that management over time [1]. Certifying a management system is a form of third-party conformity assessment [3].
Where it stops. The certification concerns the AI management system within its stated scope. It can include examination of how the organisation manages AI systems, but does not by itself establish every deployed system's accuracy, safety, legal compliance or freedom from loss. A management-system certificate is not a performance guarantee for each AI system.
2. "Our certificate is accredited"
What it establishes. Accreditation is a check on the certification body, not on the organisation it certified. UKAS's programme for AI management system certification refers to ISO/IEC 17021-1, the general standard for certification bodies, and to ISO/IEC 42006:2025, which adds requirements for bodies certifying AI management systems [2]. UKAS reported its first accreditation for ISO/IEC 42001 certification, granted to BSI, in January 2026 [4].
Where it stops. Accreditation demonstrates that the certification body has been assessed for competence and impartiality in the activities covered by its accreditation. It strengthens confidence in the certification process without widening the customer's certified scope. Certification can also be issued without accreditation. UKAS advises checking the certifier's accreditation and its scope [5]. Since 1 January 2026, Global ACI has operated the international recognition arrangement that brings together the former IAF and ILAC arrangements; the relevant recognition scope also matters [6].
3. "Our certificate covers the business"
What it establishes. A certificate applies to the scope stated on it, such as named legal entities, sites, activities or systems.
Where it stops. The certificate does not cover entities or activities outside its stated scope. A subsidiary, an acquisition or a new product may fall outside it. That does not establish whether it has been assessed separately. UKAS advises confirming that the certificate itself is valid with the certification body that issued it [5]. IAF CertSearch can help verify certificate information. IAF MD 28:2023 introduced mandatory uploading of certification data within its scope from 26 October 2024. A missing database result should be investigated with the issuer; it does not, by itself, establish that a certificate is invalid [5, 7].
4. "We were certified in 2025"
What it establishes. Certification follows an audit and a certification decision. The usual ISO/IEC 17021-1 programme starts with an initial audit, continues with surveillance and returns to recertification within a three-year cycle. A public explanation of the relevant clauses by an accreditation body describes surveillance in the intervening years, with the first surveillance audit no later than 12 months after the initial certification decision [8].
Where it stops. A certificate's issue date need not be its latest audit date. Neither date alone establishes when an AI system was most recently reviewed. Systems can change, and certification can expire, be suspended or have its scope changed. Current certification is not continuous confirmation of every system's behaviour [5, 8].
5. "We have a SOC 2 report"
What it establishes. A licensed certified public accountant (CPA) firm has examined controls for the service organisation's system in the trust-services categories covered by the report. A Type 1 report addresses the system description and the suitability of control design at a specified date. A Type 2 report also addresses whether the controls operated effectively during a specified period [9, 10].
Where it stops. A Type 1 report does not give an opinion on operating effectiveness over a period. A Type 2 report concerns the period stated in it, not later operation. The auditor's conclusion, any exceptions, exclusions of third-party services and relevant customer responsibilities can all affect what a reader may infer. The AICPA describes SOC 3 as a less detailed general-use report that can be freely distributed; SOC 2 provides more detail, and its intended users and conditions of use matter. These reports address the controls and categories within their scope, not an AI system's overall performance [9, 10, 11].
6. "We are aligned with the NIST AI Risk Management Framework"
What it establishes. The organisation is claiming to use NIST AI RMF 1.0 to guide its AI risk management. NIST published version 1.0 in January 2023 and presents it as a framework organisations may choose to use [12].
Where it stops. The word "aligned" alone does not establish how extensively the framework has been applied or whether an independent party has assessed that application. Supporting internal evidence may exist. Any independent assessment has its own scope and conclusions; the alignment claim is not itself a NIST certification. NIST reports that revision of version 1.0 is in progress [12].
7. "Our system carries the CE marking under the EU AI Act"
What it establishes. Where the AI Act's CE-marking requirements apply, the provider uses the marking to indicate conformity and assumes responsibility through its EU declaration of conformity. The applicable conformity-assessment procedure must be followed before the high-risk system is placed on the market or put into service [13].
Where it stops. The assessment route depends on the system. Article 43(2) specifies internal control without a notified body (an authorised conformity-assessment body) for the high-risk systems in Annex III points 2 to 8. Article 43(1) provides routes for the systems in point 1. Systems also covered by the product legislation in Annex I Section A follow the relevant sectoral procedure under Article 43(3). Article 43(4) requires a new assessment after a substantial modification of an already assessed high-risk system, while distinguishing predetermined learning changes documented at the initial assessment. CE marking is not an independent performance rating or, on its own, proof of compliance [13].
Applicability is phased and was amended in 2026. The European Commission's AI Act Service Desk identifies its text as based on the consolidation of 27 July 2026, and its Article 113 shows 2 December 2027 for specified Annex III high-risk provisions and 2 August 2028 for specified Annex I provisions. Those dates do not apply uniformly to every article. The applicable provisions and transitional rules need to be checked in the authentic text for the system concerned before this description is relied on [13].
8. "Our model passed an independent audit"
What it establishes. The claim is that an independent party assessed the model or system against a stated basis. The report is needed to establish what was assessed, by whom and with what conclusion. The UK government's introduction to AI assurance sets out several kinds of work that can sit behind such a claim, among them reviews of whether a system's data and outcomes are unfairly biased, reviews of adherence to policies and rules, forward-looking assessments of a system's wider effects, and mathematical proof that a system meets stated requirements [14].
Where it stops. Findings relate to the object, context, methods and scope assessed. A model test does not by itself establish how a complete deployed system will behave. Independence and a successful result cannot be inferred from the word "audit" alone. Findings do not automatically establish performance after a change; their continuing relevance depends on what changed and whether further assessment is needed. A model assessment is not, by itself, certification of the organisation [14].
What none of these claims settles on its own
The underlying certificate or report may identify its scope and relevant dates. Those boundaries need to match the business, system and decision concerned. Dated findings do not establish everything that has happened since, or guarantee future results. Insurance cover and price depend on the policy terms, the circumstances of the risk and the insurer's own decisions. Questions about the interpretation of a policy require appropriate advice.
About this page
Arkaya connects governance and cybersecurity platforms with the insurers, MGAs and lenders who price and set terms for AI and other governance risks. This page explains public assurance schemes in general terms. It does not rate, rank or endorse any scheme, certification body or organisation. Bodies are named only to identify public sources, not to imply an affiliation or endorsement. It is not legal, regulatory or insurance advice. Corrections are welcome: email djm@arkayarisk.com with the subject "Corrections: AI assurance claims reference".
© 2026 Arkaya Risk Limited. All rights reserved. You may quote short extracts with attribution to Arkaya Risk and a link to this page. Republishing the page in full, or an adapted version, needs our written permission. Linked sources belong to their owners.
Last checked: 29 October 2026.
References
- ISO, "ISO/IEC 42001:2023: AI management systems", public overview. Describes the standard as setting requirements for an organisation's AI management system. https://www.iso.org/standard/42001 (accessed 29 October 2026).
- UKAS, "Artificial Intelligence Management Systems" (development programme page), last modified 25 August 2026. Identifies ISO/IEC 42001:2023, ISO/IEC 17021-1:2015 and ISO/IEC 42006:2025 as the standards behind accredited AI management system certification. https://www.ukas.com/accreditation/about/developing-new-programmes/development-programmes/aims/ (accessed 29 October 2026).
- ISO, "ISO/IEC 17021-1:2015 Conformity assessment: Requirements for bodies providing audit and certification of management systems, Part 1", public catalogue abstract. Classes management system certification as third-party conformity assessment. https://www.iso.org/standard/61651.html (accessed 29 October 2026).
- UKAS, "UKAS grants first accreditation for ISO/IEC 42001", news item dated 15 January 2026. Reports BSI as UKAS's first accredited certification body for AI management systems. https://www.ukas.com/resources/latest-news/ukas-grants-first-aims-accreditation/ (accessed 29 October 2026).
- UKAS, "Validating management system certificates", guidance page. Sets out checks on the accreditation body, the certification body's accreditation scope and the validity of the certificate with its issuer. https://www.ukas.com/accreditation/about/validating-ms-certificates/ (accessed 29 October 2026).
- Global ACI, "Q&As" and "About Global ACI". Explain the international recognition arrangement operated from 1 January 2026, bringing together the former IAF and ILAC arrangements. https://global-aci.org/en/faq/ and https://global-aci.org/en/about-global-aci/ (accessed 29 October 2026).
- International Accreditation Forum, IAF MD 28:2023, and UKAS technical bulletin "Activation of CertSearch account", 7 October 2024. Set out the duty to upload certification data within scope, applying from 26 October 2024. https://iaf.nu/iaf_system/uploads/documents/IAF_MD_28_Issue1_26102023.pdf and https://www.ukas.com/resources/technical-bulletins/activation-of-certsearch-account/ (accessed 29 October 2026).
- International Accreditation Service (IAS), "ISO/IEC 17021-1:2015, Section 9: Process Requirements", public training explanation, PDF page 2, addressing clauses 9.1.3.2 and 9.1.3.3. An accreditation body's explanation, not the normative ISO text. https://cdn-v2.iasonline.org/wp-content/uploads/2021/02/17021-1-2015-Section-9-1.pdf (accessed 29 October 2026). See also reference 3.
- AICPA, testimony to the US Department of Labor ERISA Advisory Council, "Cybersecurity Issues Affecting Health Benefit Plans", 18 July 2022, page 6, "SOC 2 Criteria". Explains the difference between Type 1 and Type 2 SOC 2 reports. https://www.aicpa-cima.com/resources/download/testimony-on-cybersecurity-issues-affecting-health-benefit-plans and https://www.dol.gov/sites/dolgov/files/EBSA/about-ebsa/about-us/erisa-advisory-council/aicpa-testimony-on-cybersecurity-services.pdf (accessed 29 October 2026).
- AICPA, public SOC 2 report review resource. Explains the points a user of a SOC 2 report considers, including scope, period, the auditor's conclusion, exclusions and customer responsibilities. https://www.aicpa-cima.com/resources/download/soc-2-report-review-checklist and https://assets.ctfassets.net/rb9cdnjh59cm/3xbcLlNc5rd72So4nQpNIk/7a3e8e5945c78c35fc5e116859b96e6a/SOC_2_Report_%C3%82_Review_Checklist.pdf (accessed 29 October 2026).
- AICPA, "SOC 3: SOC for Service Organizations: Trust Services Criteria for General Use Report". Describes SOC 3 as covering the same categories as SOC 2 with less detail, for general use and free distribution. https://www.aicpa-cima.com/topic/audit-assurance/audit-and-assurance-greater-than-soc-3 (accessed 29 October 2026).
- NIST, "Artificial Intelligence Risk Management Framework (AI RMF 1.0)", NIST AI 100-1, publication record dated 26 January 2023, with the AI RMF overview and FAQs. Describe a framework that organisations can adapt when managing AI risks, and report that revision is in progress. https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10, https://www.nist.gov/itl/ai-risk-management-framework and https://www.nist.gov/itl/ai-risk-management-framework/ai-risk-management-framework-faqs (accessed 29 October 2026).
- European Commission AI Act Service Desk, Articles 16 (provider obligations), 43 (conformity assessment), 47 (EU declaration of conformity), 48 (CE marking) and 113 (entry into force and application), Regulation (EU) 2024/1689, based on the consolidation of 27 July 2026. Non-binding presentation; the authentic text is on EUR-Lex. Article 16, Article 43, Article 47, Article 48 and Article 113 (accessed 29 October 2026).
- Department for Science, Innovation and Technology, "Introduction to AI assurance", 12 February 2024. Describes AI assurance techniques, including bias audit, compliance audit, impact assessment, formal verification and conformity assessment, and their use across a system's life. https://www.gov.uk/government/publications/introduction-to-ai-assurance/introduction-to-ai-assurance (accessed 29 October 2026).