ISO/IEC 27090:2026 is ISO and IEC guidance on the security threats specific to artificial intelligence systems, and how to detect and mitigate them across the AI life cycle. Published in October 2026, it is guidance rather than a certifiable standard, and it works with ISO 27001 and ISO 42001 rather than replacing them.
What does ISO 27090 cover?
It covers the security threats that are specific to AI systems, not general IT security. ISO/IEC 27090:2026, published on 14 October 2026 by the joint ISO and IEC committee JTC 1/SC 27, is a 56 page guidance document. Its scope is the security threats and compromises that affect AI systems across their life cycle, from the data used to train a model through to its operation and later updates. It explains how those threats can be exploited, what the consequences can be, and which measures help detect and address them. It applies to organisations of any size and type that develop or use AI, including private companies, government entities and not for profit bodies. The standard is clear that established security practice still matters: it adds an AI specific layer on top of, not instead of, the controls an organisation already runs.
Is ISO 27090 a certification?
No. ISO 27090 is a guidance document, so there is no certificate and no audit against it. This is the most common misunderstanding, so it is worth stating plainly: an organisation cannot get certified to ISO 27090 the way it can with ISO 27001 or ISO 42001. Guidance standards describe good practice and help teams make decisions. Requirements standards, the ones written with “shall” clauses, are what accredited bodies certify against. If an organisation wants a certifiable position on AI, that is ISO 42001, the AI management system standard. ISO 27090 is the reference you use to understand and treat AI specific threats, either inside that management system or within an existing security programme. For how the two relate, see ISO 27090 vs ISO 42001 and is ISO 27090 mandatory or certifiable.
What AI threats does ISO 27090 address?
It groups AI compromises by the asset affected and the security property at risk. The threats it describes include data poisoning, where manipulated training data changes how a model behaves; evasion attacks, where crafted input fools a model at inference; membership inference, which tests whether a specific record was in the training data; model inversion and model theft, which reconstruct or lift the model itself; training data leaks and sensitive model output; and prompt injection, where crafted instructions trigger unintended behaviour. Many of these are familiar from the OWASP Top 10 for LLMs and from data poisoning research. ISO 27090 collects them in one place, ties each to its likely consequence, and points to detection and mitigation. The standard sets this out in its Clause 6; ISO 27090 threats and mitigations walks through the categories.
How does ISO 27090 work with ISO 27001 and ISO 42001?
It fills the AI specific gap between them. ISO 27090 says established information security controls still apply to AI systems, and it points to ISO 27001 and ISO 27002 for that baseline. It then adds the AI specific threat knowledge those general standards do not carry. ISO 42001, the AI management system standard, gives the governance wrapper: policies, roles, an AI risk and impact assessment, oversight. ISO 27090 feeds that governance with concrete threats and mitigations to assess against.
| Standard | Question it answers | Certifiable |
|---|---|---|
| ISO 27001 | Is our information secure? | Yes |
| ISO 42001 | Do we govern our AI responsibly? | Yes |
| ISO 27090 | What attacks are specific to AI, and how do we treat them? | No (guidance) |
The standard also references AI risk management guidance in ISO 23894 and complements the NIST AI Risk Management Framework; see ISO 27090 vs the NIST AI RMF.
Who should use ISO 27090?
Anyone who builds, runs, buys or assures AI systems. The standard names security teams, AI developers, data scientists, software engineers, risk managers, governance teams, and procurement or assurance teams checking a supplier’s controls. In practice three groups get the most from it: teams developing AI products, who need to secure training data and models; teams deploying third party or hosted AI, who need to know what to ask a vendor; and risk or board readers who need to understand where AI changes the organisation’s exposure. The standard notes sectors where an AI failure can cause serious harm, including healthcare, robotics and automotive. For Australian organisations already moving on AI governance, it pairs with the work in AI governance for Australian business.
What does ISO 27090 mean for Australian organisations?
It gives a common reference point just as AI risk becomes a board issue here. Australia has no law requiring ISO 27090, and none is expected, because it is guidance. Its value is practical: a credible, vendor neutral list of AI specific threats you can assess your own systems and your suppliers against, published by the same committee behind ISO 27001. That matters now, because the ACSC 2025-26 threat report names AI as the defining force in the threat environment, and because ASD’s AI considerations guidance is pushing AI security up the agenda. If you are pursuing ISO 42001 or building an AI governance programme, ISO 27090 is the threat library you treat against. Cybernion helps organisations stand up that governance through ISO 42001 readiness.
Frequently asked questions
Is ISO 27090 mandatory?
No. It is guidance, not law, and it is not mandated in Australia. Its value is as a credible, vendor neutral reference for AI specific threats.
Can you get certified to ISO 27090?
No. It is a guidance document with no certification scheme. The certifiable AI standard is ISO 42001, the AI management system standard.
When was ISO 27090 published?
14 October 2026, as Edition 1, by the joint ISO and IEC committee JTC 1/SC 27. It runs to 56 pages.
Does ISO 27090 replace ISO 27001 or ISO 42001?
No. It complements them. ISO 27001 secures information, ISO 42001 governs AI, and ISO 27090 addresses the threats specific to AI.
Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.
If you are building or buying AI and want to get ahead of these risks, talk to us.
Sources:
- ISO/IEC 27090:2026, Cybersecurity, Artificial intelligence, Addressing security threats and compromises to AI systems, October 2026
- ISO/IEC 27090:2026 catalogue entry, iTeh Standards, 2026
- ISO/IEC 42001:2023, Information technology, Artificial intelligence, Management system, December 2023
Last updated: 27 October, 2026
