ISO 42001 in 2026: Which AI Vendors Are Certified and What Certification Costs Your Company

AI governance has its own certificate now, and enterprise buyers have started asking for it. Only about 220 organisations worldwide have publicly announced ISO/IEC 42001 certification, and 113 of those arrived in 2026. Anthropic, OpenAI, Google, Microsoft and AWS hold it. Mistral and Manus do not show it publicly. This guide covers what the standard demands, who has it, what certification costs and how it fits with the EU AI Act.

What ISO/IEC 42001 is and what it requires

ISO/IEC 42001:2023 is the first international standard for an AI management system (AIMS) that an accredited body can certify. It follows the same structure as ISO 27001 for information security and ISO 9001 for quality: you define a scope, set policies, assign responsibility, assess risks, run controls, measure, and improve. What is new is the subject matter. The standard and its Annex A controls cover:

• AI policy and roles. Who may build, buy and deploy AI, and who is accountable for each system.
• AI risk assessment and AI system impact assessment. Effects on individuals, groups and society, not only on the company.
• Data and lifecycle controls. Data quality and provenance, development, verification, deployment, monitoring and retirement.
• Third parties. Requirements for AI suppliers and for customers who use your AI.
• Transparency. Information for users and stakeholders about how a system works and where its limits are.

It is a management standard. A certificate shows that the organisation runs a working governance system. It does not show that a particular model is accurate, safe or unbiased. For the model-level safety commitments of the big labs, see our overview of frontier AI safety frameworks.

Adoption: from 2 certificates to 220 in two years

There is no official register of ISO 42001 certificates. The best public count is the ISO42k directory, which tracks announced certifications. As of 11 October 2026 it lists 220 organisations in 29 countries, audited by 26 certification bodies: 19 announced in 2024, 87 in 2025 and 113 so far in 2026. New certifications have run at 33 to 41 per quarter since the start of 2026.

The count is small for a reason. ISO 27001 has been around for almost two decades, while ISO 42001 was only published at the end of 2023 and accredited certification bodies needed most of 2024 to get approved. The sector split shows who moved first: enterprise software companies make up 44% of certified organisations, professional services 15%, cloud and AI platforms 9%, cybersecurity 8% and GRC and compliance vendors 5%. These are companies whose customers send security questionnaires.

Where certified companies are based

Two out of three announced certificates belong to US companies, which reflects how US enterprise procurement works: SOC 2 reports and certificates are a ticket to the deal. Europe is barely visible. Only the United Kingdom (9) and Spain (5) make the top eight, and Germany, Austria and Croatia sit in the long tail. For a Croatian or DACH software or services company, that means a certificate is still a real differentiator in tenders and enterprise sales, and not yet a standard requirement.

Which AI vendors are certified: Anthropic, OpenAI, Mistral, Manus and others

For most companies the more pressing question is not whether to certify themselves, but whether their AI suppliers are certified. A vendor's ISO 42001 certificate is evidence you can file in your own supplier assessment, and it covers a large part of what your AI vendor selection questionnaire asks. The status as of mid-October 2026:

Sources: Anthropic news (13 Jan 2025); AI Provider Trust Registry checks of vendor trust pages, 12–16 Oct 2026; Manus security page. "No public evidence" means no certificate could be confirmed from public sources, not that the vendor has none. Ask the vendor for the certificate and its scope before you rely on it.

For European buyers the result is a little awkward. Mistral is the vendor with the strongest European data-sovereignty story, and it already has ISO 27001, ISO 27701 and SOC 2, which cover security and privacy. What it does not show publicly is a certified AI management system, while its US competitors do. Manus, the agent platform, is in the same position. Neither gap is a reason to rule a vendor out, but procurement and compliance teams will ask about it, so it is worth requesting a written statement on AI governance controls.

Who audits: the certification bodies

Schellman, which also certified Anthropic, has issued almost a third of the known certificates, and A-LIGN another 12%. Both are US audit firms. The European certification bodies SGS, BSI, DNV and AENOR together account for 26% of known certificates. For a company in Croatia, Germany or Austria this matters in practice: an auditor with local staff, local-language audits and accreditation recognised by your customers usually makes the process cheaper and faster. Always check that the body is accredited for ISO 42001 (for example by ANAB, UKAS or DAkkS). A certificate from an unaccredited issuer carries little weight in a tender.

What certification costs and how long it takes

The audit itself is the smaller line item. Published estimates put the certification body's fee for Stage 1 and Stage 2 at roughly $5,000 to $50,000, depending on scope and auditor, followed by surveillance audits every year. The larger cost is preparation: inventorying AI systems, writing the policy and impact assessments, setting up controls and, above all, running the system for a few months so there is evidence to audit. Consultancies put a mid-sized programme at around $45,000 to $130,000 in the first year, with most organisations needing 4 to 9 months. Companies that already run ISO 27001 typically take 4 to 7 months, because the management-system scaffolding (document control, internal audit, management review) already exists.

• Start with an AI inventory. You cannot scope an AIMS without knowing which AI systems you build, buy and use, including the ones employees adopted on their own. Our shadow AI governance guide shows how to find them.
• Keep the scope tight. Certify one product line or the AI you provide to customers first. A smaller scope means fewer audit days and a faster first certificate.
• Reuse ISO 27001. Integrate the AIMS into an existing ISMS instead of building a second system next to it.
• Collect supplier evidence. Your vendors' ISO 42001, ISO 27001 and SOC 2 reports feed directly into your third-party controls.

ISO 42001 and the EU AI Act

ISO 42001 is voluntary and does not give a presumption of conformity with the EU AI Act. The European Commission found that ISO 42001's goals and definitions are not aligned with the quality management system that Article 17 of the AI Act requires from providers of high-risk systems. That is why CEN-CENELEC JTC 21 wrote a dedicated European standard, EN 18286, published in July 2026. It has not yet been cited in the Official Journal, so it does not yet give a legal presumption of conformity either.

The timeline has also moved. The Digital Omnibus, in force since 27 July 2026, pushed the obligations for stand-alone high-risk AI systems under Annex III (employment, credit, education, biometrics and similar) to 2 December 2027, and those for AI embedded in regulated products under Annex I to 2 August 2028. Transparency duties under Article 50 have applied since August 2026. Our EU AI Act compliance guide has the full timeline.

Who should certify, and who should not

The bottom line

ISO 42001 has gone from a niche standard to a procurement checkbox in two years, but with about 220 public certificates it is still early, especially in Europe. When choosing AI vendors, ask for the certificate and its scope: Anthropic, OpenAI, Google, Microsoft and AWS can provide one, while Mistral and Manus currently rely on ISO 27001, ISO 27701 and SOC 2. If you sell AI-based products yourself, a certificate in 2027 will open doors, and building the management system now prepares you for the AI Act deadlines that follow.