What is Data Drift in AI Governance?
Data drift is a change over time in the data a system sees, or in the decisions it makes, so that the rules it was checked against no longer describe what it does. It is found by comparing current decisions with the sealed rules.
Why it matters
Rules sealed against last year's applicants may not fit this year's. Drift is not always a fault, but it is always something to decide on: update the rules, or bring decisions back within them.
How NOMOS shows it
Studio compares new decision logs with the sealed rules, threshold by threshold, and reports where they have moved apart and whether the change affected outcomes. The Registry's scheduled check does the same for policies in force, and drafts a proposed change for people to review when the evidence is strong.
Example
A sealed affordability rule caps debt obligations at 40% of income.
Over a year, approvals creep up to 46% as volumes rise. Nobody notices until losses do.
The comparison with new decision logs shows the threshold in practice has moved to 46%, and whether defaults changed with it.
Frequently asked questions
Related concepts
See it in practice
Upload a policy, confirm its rules, seal them: any system can then check an action against them.