ISO/IEC AI Standards Timeline: Publication Dates and Revisions to Track

Why a dated reference matters for AI standards

ISO/IEC JTC 1/SC 42 produces the AI standards that regulated enterprises most often cite in control frameworks. Its outputs are not one kind of document. Some are management-system standards, some are risk guidance, some are life cycle or application guidance, some are testing specifications, and some set requirements for auditors and certification bodies. A change to one of them has different consequences from a change to another.

The operational problem is edition drift. A control that was mapped to one edition, or to a draft, can quietly become out of date. When an auditor asks which edition governed AI operations on a given date, the answer needs to come from a dated record, not from memory.

The timeline below uses the designation, edition and status shown in the SC 42 catalogue as of November 2026. It records editions and statuses only. Catalogue status does not establish conformity or certification outcomes for any organisation.

SC 42 editions, amendments and work items by date

Dates here are edition years or catalogue page dates, not necessarily exact publication dates. Record the label and the date you checked it.

• 2022: ISO/IEC 22989:2022 and ISO/IEC 23053:2022: Published base editions. Amendment 1, Generative AI, is listed for each at the FDAmd 1 stage in the catalogue record updated 3 November 2026. The ISO/IEC 23053 page describes its amendments as under development.
• 2023: ISO/IEC 23894:2023: AI guidance on risk management. Published, Edition 1. The ISO page was last updated 30 October 2026.
• 2023 and 2024: ISO/IEC 5338:2023 and ISO/IEC 5339:2024: AI system life cycle processes and guidance for AI applications, both listed in the SC 42 catalogue.
• 2025: ISO/IEC 42006:2025 and ISO/IEC TS 42119-2:2025: Requirements for AIMS audit and certification bodies, and an overview of testing AI systems. The TS 42119-2 page is dated 3 November 2025, one day before the 12-month window ending 4 November 2026.
• 2026: ISO/IEC 27090:2026: Cybersecurity guidance for artificial intelligence. Published, Edition 1. The ISO page is dated 14 October 2026.
• October 2026: ISO/IEC 42105: Guidance for human oversight of AI systems. Shown as Under publication on a page dated 22 October 2026.

What changed in the previous 12 months

Two signals stand out in the catalogue for the period ending November 2026. The first is the publication of ISO/IEC 27090:2026 as an Edition 1 standard, which gives security teams a dedicated AI cybersecurity reference. The second is the movement of ISO/IEC 42105 to Under publication, which indicates human oversight guidance is close to becoming a published reference.

The third item is a work item rather than a release. Amendment 1 on generative AI is listed against both ISO/IEC 22989:2022 and ISO/IEC 23053:2022. Because the 23053 page and the later catalogue record describe the stage differently, treat the most recent official record as authoritative and confirm the stage each time you check. Until an amendment is published, it should not be mapped as final content.

ISO/IEC 23894:2023 remains a published Edition 1 standard, and its page was refreshed in October 2026. A page refresh is not a new edition, so the register should note the check date without changing the edition field. ISO/IEC 42006:2025 and ISO/IEC TS 42119-2:2025 fall outside or at the edge of the window, but they remain relevant for audit and testing scope.

What a change means for control mapping and evidence

The consequence of a change depends on the standard family. A management-system standard such as ISO/IEC 42001 defines requirements that an organisation's management system is assessed against, so a new edition can alter certification scope and the evidence an auditor expects. ISO/IEC 42006 sits beside it by setting requirements for the bodies that audit and certify, which affects how assessments are conducted rather than what your controls must do.

Risk guidance such as ISO/IEC 23894 and impact assessment guidance such as ISO/IEC 42005 shape how you structure risk and impact processes. A new edition usually means revisiting process documentation, criteria and the records those processes produce. Guidance and technical specifications describe good practice and do not carry the same weight as requirements standards, so your mapping should say which kind of document each control references.

Technical and life cycle standards such as ISO/IEC 5338, 5339 and TS 42119-2 influence engineering and testing practices. ISO/IEC 27090 does the same for security. Amendments add content that is not part of the base standard text, so an amendment should be recorded as its own item next to the base edition. The generative AI amendments to 22989 and 23053 are the clearest current example, because definitions and framework language that your controls cite may change once they are published.

Build a version-tracking process

A version register is a small control with a large audit benefit. It lets you show, for any date, which edition of each standard your AI controls referenced and what evidence existed under it. Keep it simple enough that it is actually maintained.

• Record the full designation: Use the complete identifier, such as ISO/IEC 22989:2022, with any Amd 1 reference, rather than the bare number.
• Capture status and check date: Store the catalogue status label and the page's last-updated date every time the entry is reviewed.
• Separate published from unpublished: Mark items that are Under publication, at FDAmd or under development, and do not map controls to them as if they were final.
• Keep families distinct: Group management-system, risk, impact, testing, oversight, security and certification-body standards separately, and flag guidance versus requirements.
• Assign an owner and triggers: Name one owner for monitoring the SC 42 catalogue and approving mapping changes. Trigger review on any new status or amendment.
• Snapshot catalogue pages: Keep dated copies of the pages you relied on as evidence of which edition was current when AI operations ran.

Linking standard editions to runtime policy and evidence

A register tells you what you are mapping to. It does not show what your AI systems actually did. For that, each tracked edition needs a link to the policy that was enforced and the evidence that behavior was consistent with it, so that an auditor can follow the chain from standard edition to control to recorded activity.

Trussed enforces policy in the execution path for models, agents, tools and MCP servers, with actions such as allow, redact, block, rate limit and escalate for human approval at the level of individual interactions and tool calls. That enforcement generates audit-ready evidence of what the AI was allowed to do and what it did. Your register supplies the other half, by recording which standard edition each policy was written against. Trussed does not determine certification or conformity outcomes.