Equilibrium Fairness: a runtime governance architecture for monitoring and correcting fairness drift

Today I published a working paper titled: Equilibrium Fairness: A Runtime Governance Architecture for Monitoring and Correcting Fairness Drift in High-Impact AI Systems.

The paper explores an implementation gap that sits between AI governance frameworks and operational practice. While frameworks such as the EU AI Act, ISO/IEC 42001 and the NIST AI RMF establish important obligations around oversight, monitoring and accountability, organisations still need practical runtime mechanisms that connect those obligations to deployed systems.

The paper proposes a governance architecture built around four stages: initialise, monitor, threshold and correct, supported by a separation-of-roles principle in which monitoring may be automated but corrective action remains human-governed.

This is a working paper and an invitation to discussion, critique and collaboration.

Read the paper: Equilibrium Fairness on Zenodo (DOI: 10.5281/zenodo.20547396)

If you are working on AI assurance, fairness monitoring, or human oversight and would like to discuss or collaborate, get in touch.