Technical Guide

    ISO/IEC 5339 Guidance for AI Applications and Stakeholders

    ISO/IEC 5339:2024 is guidance for AI applications that emphasizes stakeholder engagement and the AI application life cycle. It is not a certifiable requirement set. Enterprise architects can use it to decide who an application serves and affects, in what context it runs, and which of those concerns can be enforced as runtime policy and recorded as durable evidence.

    Talk to an Expert See How Evidence Is Produced

    Translating the guidance into the execution path

    1. 1

      Stakeholder perspectives become policy subjects

      Attribute each decision to the Make, Use or Impact concern and to a named owner.

    2. 2

      Context of use becomes policy conditions

      Express clause 5.2 context as application, model, agent, tool or MCP server, data class and user.

    3. 3

      Impact concerns become enforceable rules

      Convert a small set of impact concerns into redact, block or rate limit rules.

    4. 4

      Non-functional characteristics become attributes

      Define measurable attributes whose enforcement decisions are recorded.

    5. 5

      Escalation paths become approvals

      Route sensitive actions to a named approver and record the result.

    6. 6

      Processes map to life-cycle stages

      Tie clause 5.3.4 processes to the stage where each policy is changed and reviewed.

    Runtime policy and audit-ready evidence checklist

    • List each AI application's stakeholders across the Make, Use and Impact perspectives, and assign a named owner to each.
    • Record the context of use and the life-cycle processes for each application in a form that policy can reference.
    • Reduce contextual and impact concerns to a short set of enforceable rules with defined outcomes: allow, redact, block, rate limit, escalate.
    • Record every decision with timestamp, policy version, subject, resource and outcome, plus the approver for any escalation.
    • Keep evidence durable, exportable and retained for a period you define, with data residency that matches your obligations.
    • Map controls to GOVERN, MAP, MEASURE and MANAGE using the published crosswalk, and describe the work as guidance-informed, not conformant.

    What ISO/IEC 5339 covers, and what it does not

    The official title is ISO/IEC 5339:2024, Information technology, Artificial intelligence, Guidance for AI applications. ISO describes it as guidance that emphasizes stakeholder engagement and the AI application life cycle. Its clauses address the context of an application (clause 5.2), the impact on stakeholders (clauses 5.3.2 and 5.3.3) and processes (clause 5.3.4).

    The word guidance matters. The standard helps teams identify context, opportunities and processes for developing and applying AI applications. It does not give you a list of controls to pass an assessment against, and this page makes no claim of certification or conformance. Treat it as a structured way to ask the right questions about an application, then make your own architectural decisions.

    As of November 2026, ISO lists ISO/IEC 5339:2024 in the ISO/IEC JTC 1/SC 42 catalogue at stage 60.60, with no amendment or revision reported. Some internal documents and older references still cite earlier draft stages of the standard, so check which edition your own materials refer to.

    Stakeholder perspectives: Make, Use and Impact

    The stakeholder view in ISO/IEC 5339 is organized around three perspectives on an AI application and its context: those who make it, those who use it, and those who are impacted by it. This framing is useful because it separates accountability that is usually blurred in practice. The team that builds an agent, the business unit that operates it and the customers or patients affected by its outputs have different concerns and different authority.

    For an architect, the practical step is to turn each perspective into an explicit policy subject. Name an owner for each one. When a runtime decision is later reviewed, you should be able to say which stakeholder concern the policy served and who was accountable for it.

    This mapping is a practitioner interpretation of the standard, not text from it. Check the standard itself for the full treatment of stakeholder roles before you adopt specific role names in internal policy.

    Positioning ISO/IEC 5339 among adjacent frameworks

    The most direct, documented bridge is the crosswalk between NIST AI RMF 1.0 and ISO/IEC 5338 and 5339, hosted on NIST's AI Resource Center and submitted by the AI RMF user community in April 2024. It relates the AI RMF Core functions (GOVERN, MAP, MEASURE, MANAGE) to the stakeholder perspectives in ISO/IEC 5339. It aligns AI RMF context with clause 5.2 and processes with clause 5.3.4, and it aligns the AI RMF people and planet dimensions with stakeholder impact in clauses 5.3.2 and 5.3.3.

    The crosswalk also notes that ISO/IEC 5339 uses the term AI application non-functional characteristics for what the AI RMF crosswalk calls comparative characteristics. Practitioners can treat these as candidate runtime policy attributes.

    The practical conclusion is to avoid building a separate control set for 5339. Use ISO/IEC 5339 to frame stakeholders, context and life-cycle processes, and use the AI RMF functions to organize the activities and evidence that follow.

    Where a runtime control layer supports this work

    Trussed is a runtime control layer for securing AI. It enforces policy in the execution path across models, agents, tools, MCP servers, AI sites and coding assistants, with outcomes of allow, redact, block, rate limit and escalate for human approval. Enforcement applies to individual interactions and tool calls and happens at runtime, not after the fact.

    That enforcement produces continuous audit-ready evidence, mapped to frameworks including NIST and the EU AI Act. Trussed also provides code provenance covering the origin, licence and security context of code. It deploys self-managed in the customer cloud, with configurable retention and data residency. In production, it has run continuously for 18+ months in a regulated enterprise and holds 100+ models under one policy plane.

    Trussed does not make ISO/IEC 5339 certifiable, and it does not claim conformance to the standard. It gives regulated enterprises a way to enforce the runtime policies that their reading of the guidance produces, and to keep the evidence that shows what happened.

    From stakeholder perspective to runtime control

    Make

    Builders of the application: policy subjects, owners and change accountability.

    Use

    Operators and users: context of use expressed as policy conditions.

    Impact

    People and groups affected: concerns turned into allow, redact, block, rate limit or escalate decisions.

    Evidence

    Each decision recorded with policy version, subject, resource and outcome.

    Turn stakeholder guidance into enforced policy

    Review how policy enforcement in the execution path produces audit-ready evidence for your AI applications.

    Talk to an Expert