---
title: ISO/IEC 27090: security for AI systems, in plain English — ReadyForAI.ie
description: ISO/IEC 27090 is new international guidance on security threats specific to AI, such as data poisoning, prompt injection and model theft. How it fits with ISO 27001 and ISO 42001, and how ReadyForAI helps.
canonical_url: https://www.readyforai.ie/iso-27090
last_updated: 2026-11-13
---

# Keep your AI safe from the attacks made for it.

ISO/IEC 27090 is new international guidance on security threats that are specific to AI: poisoned data, hidden instructions, stolen models and leaking information. It isn’t something you get certified against. It’s a checklist that works with what you already have, such as ISO 27001 or ISO 42001.

## AI has its own weak spots.

Ordinary security still matters, but AI adds new ways in. ISO 27090 names them and sets out how to address them across the AI life cycle, from data to retirement.

### The kinds of threat it covers

- Poisoned data: someone slips bad examples into training data so the model learns the wrong thing.
- Hidden instructions: text in an email, web page or document tells an AI assistant to do something it shouldn’t (prompt injection).
- Fooling the model: inputs crafted to make it misread what it sees.
- Stealing or copying the model, or pulling private training data back out of it.
- Supply-chain risk: models, plug-ins and data you got from someone else.

### Who it’s for

- Anyone building AI into a product or service.
- Teams using AI assistants connected to email, files or customer data.
- Security teams with ISO 27001 who need to add AI to their risk register.
- Organisations working towards ISO 42001, which asks you to manage AI risk.
- Providers of high-risk AI under the EU AI Act, which requires robustness and cybersecurity (Article 15).

## Three standards, one picture.

1. **ISO 27001: your security system.** The information security management system many organisations already have. Certifiable.
2. **ISO 42001: your AI system.** How you govern AI across the organisation: roles, risk, impact and oversight. Certifiable.
3. **ISO 27090: the AI security detail.** The AI-specific threats and controls that plug into both. Guidance, not certifiable, so you use it as a checklist.

## What it gives you.

- **Use AI assistants with confidence.** Know what an assistant can reach and what a hidden instruction could make it do, before you connect it to your inbox.
- **Answer the questionnaire.** Customers now ask how you secure your AI. A recognised international reference makes the answer credible.
- **Evidence for the law.** Useful evidence for the AI Act’s robustness rule and the Cyber Resilience Act’s security requirements.
- **A board that gets it.** Prompt injection explained in five minutes, so directors can ask the right questions.
- **No new bureaucracy.** It slots into the risk register and controls you already run, rather than starting a new system.
- **Safer experiments.** Clear guardrails make it easier to say yes to the next AI idea.

## From AI list to AI controls.

1. **AI inventory.** What AI you use and build, what it is connected to, and who supplies it.
2. **Threat check.** Each system against the ISO 27090 threats, ranked by what could actually hurt you.
3. **Controls plan.** Fixes mapped onto your ISO 27001 or ISO 42001 controls, so nothing is duplicated.
4. **Team and board training.** What prompt injection and data leaks look like in daily work, and what to do about them.

## Make your AI hard to trick.

A 30-minute call to look at the AI you use and the threats that matter most. Guidance only, not legal advice.

- Book a call: https://calendly.com/jonathanmcguinness
- Every AI law and standard by date: https://www.readyforai.ie/ai-rulebook


---

*Part of [readyforai.ie](https://www.readyforai.ie/) · [Site map](https://www.readyforai.ie/sitemap.md) · [llms.txt](https://www.readyforai.ie/llms.txt)*
