AI Change Accountability & Reauthorisation
A practitioner-derived method for tracing AI change from signal to decision, implementation and closure.
Applied in a first organisational field pilot on a real AI change event.
AI systems do not remain fixed after approval. Models change beneath stable product names, providers retire or replace services, hosting moves, permissions widen, human review erodes, and actual use drifts from the purpose that was approved. Any of these can invalidate the assumptions behind an earlier decision — without automatically creating a new one. That gap is organisational, not technical: the change becomes visible somewhere, yet nothing returns it to a person with the authority to decide.
The method closes that gap with one practical path. A change is caught as a signal, classified, and weighed for how much it matters; it is put in front of the person or forum that actually holds the decision; and it is carried through to implemented action and a closure that is evidenced rather than assumed. The question it operationalises: which changes to an AI system, model, provider, platform, data basis, operating context, level of autonomy or human role should trigger renewed assessment, reauthorisation, restriction, suspension or exit — who has authority to decide, and how is implementation and closure evidenced?
The document is complete and self-contained: fourteen sections that carry a change from intake to closure — classification, materiality and hard-escalation triggers, review routes, authority mapping that records formal and actual authority separately, interim operating status while a decision remains open, decision, action, closure and reopen logic, an evidence-quality model and two fictional worked examples — plus six appendices of blank templates: the Organisation Authority Map, the two-part AI Change Event Sheet, the Closure & Learning Supplement, the Event Register specification and the Boundary Record.
The field evidence from the pilot is reported in a companion field study:
Kurek, M. (2026), Governing AI After Deployment: Field Evidence on Model Changes, Platform Configuration and Organisational Authority, working paper, 21 August 2026, SSRN, DOI: 10.2139/ssrn.7329378.