AI Governance
AI change control should begin before the first model update
A practical framework for connecting intended use, model changes, performance evidence, risk, and release decisions.
AI governance becomes expensive when it is treated as a final documentation exercise. The stronger approach is to define the change-control system while the product and evidence strategy are still taking shape.
Start with the claim boundary
Every assessment should begin with intended use, users, use environment, inputs, outputs, and the role of the model in the clinical or operational workflow. A change that appears small at the code level may be consequential if it changes any of these boundaries.
Build a change taxonomy
Separate changes to data, architecture, training, thresholds, user interface, workflow integration, infrastructure, and monitoring. For each class, define the evidence required, the approval authority, and the regulatory assessment route.
The unit of governance is not the model alone. It is the model, its evidence, its workflow, and its controlled use.
Operational questions
- Does the change affect intended use or performance claims?
- Could it introduce new hazards, bias, or cybersecurity exposure?
- Which verification, validation, or clinical evidence must be repeated?
- Does the authorized change protocol cover it?
Further regulatory context
Read medical device regulatory definitions and authoritative references. Explore the workflows eeRegOS supports.
Editorial note: This content is educational and does not constitute legal or regulatory advice. Product-specific decisions should be reviewed by qualified professionals using current authoritative sources.