What is ISO/IEC 24029-3?

ISO/IEC 24029-3 is the statistical-methods part of the ISO/IEC 24029 series on assessing the robustness of neural networks. It gives a methodology for selecting, applying and managing statistical methods to assess robustness properties. It is at final draft stage, not published.

What does it cover?

The public scope says it provides methodology for the use of statistical methods to assess robustness properties of neural networks. The introduction explains why: neural networks handle high-dimensional input, the input space is hard to explore exhaustively, and statistical methods explore it more widely than formal methods can.

Who is it for?

AI governance leads, model risk and quality leads, and anyone who has to show robustness evidence for a system that contains a neural network. The scope does not limit it by sector or organisation size.

What does it not do?

It does not set a threshold, a sample size or a pass mark. It does not certify, and it is not a management system standard. It does not replace ISO/IEC 42001 for governance or ISO/IEC 23894 for the risk process.

When does it publish?

ISO records stage 50.00, FDIS registered, on 2026-08-21. No publication date is recorded. Our planning date of 16 February 2027 is an estimate.