ISO/IEC 27090, EN 18286 & IEEE 2863: The 3 new AI governance standards of 2026 at a glance

Three standards in three days: why October 2026 matters for AI governance

Between 14 and 16 October 2026, three AI governance standards took effect at once: ISO/IEC 27090:2026 was published on 14 October, EN 18286:2026 was cited in the Official Journal of the EU the same day as the first harmonised standard under the EU AI Act, and IEEE 2863-2026 followed on 16 October. Three bodies, three perspectives, one topic: how do organisations build and run AI securely, compliantly and responsibly?

For organisations that develop or use AI, all three are worth a look. They cover different levels – technology, regulation and organisation – and complement rather than compete with each other. This article explains what the standards contain, who they apply to and what to do now.

The three standards at a glance

ISO/IEC 27090:2026 – Cybersecurity for AI systems

ISO/IEC 27090:2026 is the first international standard dedicated entirely to security threats against AI systems. Its full title is Cybersecurity — Artificial intelligence (AI) — Addressing security threats and compromises to AI systems. It was developed by ISO/IEC JTC 1/SC 27, the committee behind ISO/IEC 27001. ISO standard record

The focus is on protecting AI systems, not on using AI to defend against other attacks. The standard describes which attacks are specific to models, training data and AI workflows, how to detect them and how to counter them across the whole lifecycle. They include:

• Data poisoning: manipulated training data that changes model behaviour
• Model exfiltration: extracting a model through seemingly legitimate queries
• Evasion attacks, membership inference, model inversion and direct model poisoning
• Prompt injection as well as sensitive or harmful model outputs

ISO/IEC 27090 also covers cross-cutting topics such as threat modelling, red teaming, zero trust, supply chain security, logging and monitoring, and how models and safeguards age over time. Its guiding principle is defence in depth: several layers of protection, because some AI attacks are hard to detect. ISO explainer on ISO 27090

An important distinction: ISO/IEC 27090 is guidance, not a certifiable requirements standard. It adds AI-specific security aspects to an existing ISMS under ISO/IEC 27001 and an AI management system under ISO/IEC 42001. For security teams, it is the natural entry point for bringing AI risks into existing processes. Read more in ISO 27090 explained for AI security teams and ISO 27090 vs ISO 42001 explained.

EN 18286:2026 – The quality management system for the EU AI Act

EN 18286:2026 is the first harmonised standard under the EU AI Act – and so the one of the three with the greatest legal weight. CEN and CENELEC published Artificial intelligence – Quality management system for EU AI Act regulatory purposes in July 2026. On 14 October 2026, the European Commission cited it in the Official Journal of the EU with Implementing Decision (EU) 2026/2274.

That citation is the decisive step. Under Article 40 of the AI Act, providers that apply a harmonised standard published in the Official Journal benefit from a presumption of conformity for the requirements it covers. Specifically, this concerns Article 17: providers of high-risk AI systems must operate a documented quality management system. Under the updated timeline, this obligation applies from 2 December 2027 for high-risk systems listed in Annex III and from 2 August 2028 for high-risk systems in regulated products under Annex I. European Commission AI Act timeline EN 18286 turns that obligation into auditable requirements, including:

• Design and development controls
• Risk management and data management
• Documentation and records
• Post-market monitoring and incident handling

Organisations should keep three things in mind. First: applying the standard remains voluntary; conformity can be shown in other ways – just without the presumption. Second: the presumption covers only the requirements the standard actually addresses, not the whole AI Act. Third: ISO/IEC 42001 is not a substitute. The international standard for AI management systems is certifiable, but it is not a harmonised standard and does not create a presumption of conformity. Organisations that already run a 42001 system have a good foundation but must map their processes against EN 18286. Nemko Digital on EN 18286

Further harmonised standards on risk management, data quality, transparency, human oversight and cybersecurity are in progress but not yet cited. EN 18286 is currently the only AI Act reference in the Official Journal. For how to classify your AI use cases under the AI Act, read EU AI Act: classify use cases and plan evidence.

IEEE 2863-2026 – Organisational AI governance

IEEE 2863-2026 starts one level higher: with the question of how an organisation as a whole governs AI responsibly. The IEEE Recommended Practice for Organizational Governance of Artificial Intelligence was published by the IEEE Standards Association on 16 October 2026 – after more than six years of development (project start February 2020, board approval June 2026). IEEE SA standard page

The standard is aimed at an organisation's governing body – executive management, the board of directors or the supervisory board – and has three building blocks:

1. Governance principles the body should prioritise
2. Processes that put those principles into practice
3. Case studies of concrete AI applications that show how the processes implement the principles – including how to resolve conflicts between principles

The key topics IEEE names range from accountability, transparency, explainability and human oversight through privacy, safety and non-discrimination to sustainability and well-being. On the process side, they include auditing, red teaming, incident response, lifecycle oversight, value chain management, competence building and stakeholder engagement.

As a Recommended Practice, IEEE 2863 sets out recommendations, not certifiable requirements. Its value lies in bridging abstract ethical principles and concrete decision processes at leadership level. It also describes how organisations deal with external governance instruments: regulation, professional standards, codes of conduct and public expectations of responsible AI.

How the three standards fit together

The three standards address three levels that overlap in practice:

ISO/IEC 42001 is a natural common foundation. The AI management system provides the structure into which governance principles (IEEE 2863), regulatory QMS requirements (EN 18286) and security measures (ISO/IEC 27090) can be slotted. EN 18286 is structurally aligned with ISO/IEC 42001 without replacing it; ISO/IEC 27090 refers to 42001 as the management framework; IEEE 2863 describes the governance level above. Treat the three standards separately and you build the same processes three times – treat them as layers of one system and you save effort and avoid contradictions.

What organisations should do now

Priorities depend on your role in the AI value chain:

• Providers of high-risk AI in the EU: obtain EN 18286 and run a gap assessment against your existing QMS. If you already work to ISO 9001, ISO 13485 or ISO/IEC 42001, you can integrate the AI Act elements, but you must document the mapping.
• Security teams: inventory AI systems – including those sourced from third parties. Clarify where data and models come from, extend threat modelling to AI-specific attacks and check how suspicious model behaviour would be detected and handled. ISO/IEC 27090 provides the checklist.
• Executive management and supervisory boards: use IEEE 2863 as the framework for an AI governance mandate – set principles, assign responsibilities, define reporting lines and decide trade-offs explicitly.
• Everyone: watch for further harmonised standards. EN 18286 is the start; standards on risk management, cybersecurity and logging will follow.

ISO/IEC 27090 and EN 18286 are paid – ISO/IEC 27090 through the ISO Store, EN 18286 through the national standards bodies (in Germany, DIN Media). IEEE 2863 is available at no cost through the IEEE GET Program, or can be bought or accessed by subscription through IEEE SA.

Conclusion

October 2026 marks the shift in AI governance from principles to auditable standards. ISO/IEC 27090 makes AI security concrete, EN 18286 gives the EU AI Act its first harmonised standard with a presumption of conformity, and IEEE 2863 anchors AI governance at leadership level. Know all three and you can treat security, compliance and accountability as one system – rather than three separate projects.

Sources

• ISO/IEC 27090:2026 – catalogue entry and ISO/IEC 27090 explained (ISO)
• Implementing Decision (EU) 2026/2274 (EUR-Lex)
• EN 18286: EU AI Act Cites New Harmonized QMS Standard (Nemko Digital, 16 October 2026)
• IEEE 2863-2026 – standard page (IEEE SA)