ISO/IEC 27090:2026 published: ISO guidance on security threats to AI systems

ISO and IEC published ISO/IEC 27090:2026 on 14 October 2026 (first edition, prepared by JTC 1/SC 27). It gives guidance on understanding, detecting and mitigating security threats specific to AI systems across their life cycle, such as data poisoning, model theft, evasion and prompt injection, including risks introduced when models are updated or retrained on operational data. It is a guidance standard: there is no certification against it.

What it means: organisations that build or use AI can use it to extend their ISO 27001 risk assessment and Annex A controls to AI assets (training data, models, prompts, pipelines). For those working on ISO/IEC 42001, it is a practical reference for treating AI security risks. It gives no presumption of conformity with the AI Act; harmonised standards for AI Act cybersecurity requirements are still awaited.

Source: ISO, ISO/IEC 27090:2026. Read on the Cyberbits regulatory watch · Book a free 30-minute call