Unit 11. Governance, fairness and accountability

A detector that works on a test set still has to be governed. Someone must own it, its errors fall on particular people, and laws may apply to it. This unit covers the frameworks and regulation that structure those obligations as of October 2026 (the NIST AI Risk Management Framework and its generative AI profile, NIST's taxonomy of adversarial machine learning, ISO/IEC 42001:2023 and the EU AI Act with its amended timeline), the statistical criteria of group fairness and the exact conditions under which they conflict, and the documentation and audit practices (model cards, datasheets, internal audits and monitoring) that let others check a model's claims.

After the unit you will be able to place a security detector within the functions of the NIST AI Risk Management Framework and the risk categories of the AI Act, compute and compare per-group error rates with intervals, explain which fairness criteria can and cannot hold together when base rates differ, choose and evaluate per-group thresholds, and write a model card and an audit plan for a security model. The unit builds on the metrics and evaluation of Units 1 and 2, on the adversarial threats of Units 7, 8 and 10, and on differential privacy from Lesson 8.3. Plan on about 8 to 9 hours, including the lab (under a minute of running time on a laptop CPU), the exercises and the quiz.

Lessons

1. 11.1 Risk management frameworks and regulation
2. 11.2 Fairness in security decisions
3. 11.3 Model cards, datasheets and audits

Labs

Practice

• Exercises with worked solutions
• Self-check quiz