What is the difference between ISO/IEC 27090 and ISO/IEC 27091?

ISO/IEC 27090:2026 is guidance on security threats to AI systems and was published on 14 October 2026. ISO/IEC 27091 is guidance on privacy risks in AI systems and ML models, with publication estimated for February 2027. Neither is certifiable, and they are designed to be used together.

Both standards come from ISO/IEC JTC 1/SC 27 and both are guidance, so neither can be certified. They divide the AI risk landscape between them.

ISO/IEC 27090:2026 covers security threats to AI systems, such as attacks on the confidentiality, integrity and availability of models, data and pipelines. Its draft listed 13 threats. Its main readers are CISOs, security architects and ML security engineers. It was published on 14 October 2026 and refers to 27091 for privacy aspects.

ISO/IEC 27091 covers privacy risks in AI systems and machine-learning models. Its draft (DIS) lists about eight threats, including membership inference, training-data extraction and model inversion. It also covers misbehaviour such as hallucinated personal data, not only deliberate attacks. Its main readers are DPOs, privacy engineers and AI governance leads. As of 18 November 2026 it is at FDIS ballot, with publication estimated for February 2027.

Some threats matter to both. Data poisoning, for example, is a security problem that can also expose or harm individuals.

How to use them together: run one combined AI threat analysis. Record each threat once, then assess both its security impact (27090) and its privacy impact (27091). Feed the result into your ISO/IEC 42001 AI risk assessment or your ISO/IEC 27701:2025 PIMS risk assessment.

Threat counts come from the drafts. Check the published texts before you quote them.

Read more: ISO/IEC 27090 guide · ISO/IEC 27091 guide.

Explore related training

Browse all: Cybersecurity

Browse all FAQs →

Full knowledge base