Introducing ISO/IEC 27090
New international guidance on the security threats that are specific to AI systems. Published on 14 October 2026.
What is ISO/IEC 27090?
ISO/IEC 27090:2026, Cybersecurity — Artificial intelligence (AI) — Addressing security threats and compromises to AI systems, helps organisations identify, detect and reduce the security threats that are specific to AI systems, across their whole life cycle. It explains how weaknesses in AI systems can be exploited, what the consequences can be and which measures help reduce the risk.
What it covers
The standard focuses on protecting training data and AI models. The threats it addresses include:
- Poisoned training data: data altered so that a model learns the wrong behaviour.
- Model theft: a trained model copied or extracted, taking your investment and know-how with it.
- Corrupted retraining: data gathered while a system runs feeding later retraining, carrying bad data into future versions.
- Weaknesses across the whole life cycle: from design and development through deployment, operation and retirement.
How it fits with ISO/IEC 27001 and ISO/IEC 42001
ISO/IEC 27090 complements existing cybersecurity practice rather than replacing it. ISO/IEC 27001 covers information security across the organisation. ISO/IEC 42001 sets out how you govern AI as a management system. ISO/IEC 27090 goes deeper on the security of the AI systems themselves.
It is a guidance document, so certification remains against ISO/IEC 27001 and ISO/IEC 42001. ISO/IEC 27090 helps inform how AI security risks are treated within them.
Practical first steps
You can start securing your AI before reading all 56 pages:
- Include every AI system in your security risk assessment, including AI built into the software you buy.
- Know where your training data comes from, who can change it and how changes are recorded.
- Control who can access, copy or export your models.
- Put retraining and model updates under change control, with checks before anything goes live.
- Monitor AI systems in operation, and make sure your incident plan covers AI-specific attacks.
- Name an accountable owner for each AI system, with the authority to restrict or stop it.
Where to start
You cannot secure AI you do not know about. Our free 5-question AI use check takes two minutes and shows who is using AI, how often, and whether it is approved. For a fuller picture, an AI Governance Health Check gives you an independent view of where you stand and a prioritised plan.