Glossary
ISO/IEC 42001
The first certifiable management-system standard for AI, published in December 2023. It does for responsible AI what ISO/IEC 27001 does for information security: a documented system of policies, risk assessment, controls and audits.
General definition
ISO/IEC 42001 (full title ISO/IEC 42001:2023, Information technology, Artificial intelligence, Management system) defines an AI management system: the organisational structure, policies, objectives and processes through which a company governs the AI it develops or uses. It was published by ISO and IEC in December 2023 and is written for any organisation, whether it builds models, integrates them or simply deploys AI tools. It is a management standard, not a technical one: it does not tell you how to test a model, it tells you to have a documented, repeatable process for deciding what to test and acting on the result.
The standard follows the ISO harmonised structure shared by ISO 27001 and ISO 9001, with clauses 4 to 10 covering context of the organisation, leadership, planning, support, operation, performance evaluation and improvement. Annex A lists the reference controls, grouped by objective:
- AI policies and the internal organisation: roles, responsibilities and a route for reporting concerns
- Resources for AI systems: data, tooling, computing and competent people
- Impact assessment of AI systems on individuals, groups and society
- AI system life cycle: objectives, requirements, design, verification, deployment, operation and monitoring
- Data for AI systems: provenance, quality and preparation
- Information for interested parties, including documentation and incident reporting
- Responsible use of AI systems, and third-party and customer relationships
Annex B gives implementation guidance for each control, Annex C lists AI-specific objectives and risk sources (fairness, transparency, safety, security, privacy, robustness and others) and Annex D describes use across domains and sectors. Certification is carried out by accredited certification bodies through a documented audit, and the companion standard ISO/IEC 42006 sets the requirements those bodies must meet. Organisations that already run an ISO 27001 information security management system can integrate the two, since the clause structure is shared and many security controls carry over; 42001 adds the AI-specific parts: impact assessment, life-cycle governance, data provenance and transparency to the people affected.
Buyers now ask vendors for 42001 in the same breath as SOC 2 and ISO 27001. It is a useful signal that a vendor has a governed process, but it certifies the management system, not the safety of any particular model, and a certificate held by a model provider says nothing about how a downstream product uses that model.
In the Ethora ecosystem
Ethora does not hold ISO/IEC 42001 certification, nor ISO 27001 or SOC 2. The honest position is that Ethora is an open-source platform with public code, a published DPA and a described security architecture, and that customers who need a certified management system run their own programme on top of it. What the platform contributes is the operational evidence a 42001 programme asks for: an inventory of AI agents with their prompts, models and knowledge sources visible in the admin panel; an audit log and compliance trail with immutable export for the monitoring and incident-reporting controls; and configurable retention and per-user erasure for the data controls.
The life-cycle controls are easier to satisfy when the model is yours to pin. Per-agent model choice and support for self-hosted models through an OpenAI-compatible endpoint let you fix a model version, evaluate it, record the result and change it through your own approval process, rather than inheriting whatever a hosted API changed overnight. That links 42001 to the broader AI governance programme and to risk frameworks such as the NIST AI RMF, which many organisations use alongside it.