Practical guide
How to operationalize regulatory change management in AI workflows
Operational regulatory change management moves an authoritative change through a controlled queue: capture the source, decide applicability, translate it into testable requirements, coordinate implementation, verify the running state, and keep exceptions visible. A legal alert, policy update, or completed ticket is not the workflow by itself. Use CLOCK: Capture, Link scope, Operationalize.
Operational regulatory change management moves an authoritative change through a controlled queue: capture the source, decide applicability, translate it into testable requirements, coordinate implementation, verify the running state, and keep exceptions visible. A legal alert, policy update, or completed ticket is not the workflow by itself. Use CLOCK: Capture, Link scope, Operationalize requirements, Confirm implementation, and Keep under review. This is an operating procedure template, not legal advice or evidence of conformity.
Control objective
Relevant changes are identified early enough to assess their effect on AI-enabled work, assigned to authorized owners, translated into controlled changes, tested against approved requirements, and retained with exceptions and follow-up evidence.
Workflow roles
| Queue role | Authority and action | | --- | --- | | source monitor | captures official changes and monitoring failures | | legal or regulatory owner | decides interpretation, scope, and need for advice | | change coordinator | controls record, dates, dependencies, and status | | operational owner | changes the affected product, process, contract, or guidance | | specialist reviewer | reviews privacy, security, data, records, workforce, or vendor effects | | independent tester | challenges critical requirements against the implemented state | | governing owner | approves material risk, exceptions, escalation, and closure |
The NIST AI RMF Core calls for documented legal and regulatory requirements, clear roles, ongoing monitoring, and defined responsibility for AI risks. It is voluntary guidance, not a source of legal obligations. NIST AI RMF Core
Sources: [1]
C: Capture the official change
Create a queue item within the monitoring service level. Required fields are:
Secondary alerts and AI summaries may help discovery, but the queue item must link to the official source. Monitoring failures create incidents or exceptions rather than silent gaps.
The FCA Handbook provides consolidated rules, legal instruments, timeline comparison, track changes, and machine-readable current content. The FCA says its API can support compliance monitoring and regulatory or policy change management, while noting that the API does not provide historic versions. That makes version retention inside the organization's workflow important. FCA Handbook FAQ
- issuing authority and stable source;
- instrument, article, rule, guidance, or notice identifier;
- authoritative text version and retrieval time;
- publication, adoption, application, and effective dates where different;
- jurisdiction, entity, activity, system, data, and affected-person scope;
- amendment, correction, transition, and dependency information;
- monitor, backup owner, and next check.
Sources: [2]
L: Link the change to current scope
The legal or regulatory owner assigns one disposition by the triage deadline:
Record the reasoning:
| Scope field | Evidence | | --- | --- | | entity and jurisdiction | current legal and operating facts | | activity and user | process and journey map | | AI system and components | inventory, vendor, model, data, prompt, retrieval | | obligation or expectation | exact official source anchor | | current state | policy, control, contract, or behavior today | | gap and consequence | difference, reach, timing, reversibility | | uncertainty | disputed interpretation or missing fact | | decision | owner, approver, date, and reopen trigger |
The consolidated EU AI Act is a concrete example of why scope, operator role, system category, and application date must be tested separately. Review the current official text for the actual case. EUR-Lex: Regulation (EU) 2024/1689, consolidated text
- not applicable;
- monitor for a defined trigger;
- assess in detail;
- implement;
- seek advice;
- escalate immediately.
Sources: [3]
O: Operationalize the interpretation
Translate the approved interpretation into one or more testable requirements:
| Requirement | Source anchor | Affected artifact | Expected behavior | Owner | Due date | Test | | --- | --- | --- | --- | --- | --- | --- | | | | UI / policy / data / contract / model / support | | | | |
Create linked work for every affected layer. A typical AI workflow may require changes to user communication, access, data handling, model or prompt configuration, retrieval sources, vendor terms, human review, records, monitoring, support, training, incident response, and decommissioning.
Maintain the legal interpretation as a separate approved artifact. Operational teams may clarify implementation details but must not silently change the interpreted requirement. New uncertainty returns to the legal or regulatory owner.
C: Confirm implementation with evidence
Before closure:
Use a closure packet:
| Stage | Artifact | Reviewer | Result | Gap or exception | | --- | --- | --- | --- | --- | | source | frozen official text | | | | | scope | applicability decision | | | | | requirement | approved translation | | | | | implementation | changed version | | | | | test | expected and observed result | | | | | operation | current-state sample | | | | | follow-up | monitoring and review | | | |
- Freeze expected test results and the implementation version.
- Test ordinary, edge, failure, and out-of-scope conditions proportionate to risk.
- Confirm the running state matches the reviewed code, configuration, content, contract, or procedure.
- Trace each critical result back to requirement and official source.
- Keep failed, incomplete, and disputed results visible.
- Verify communication, support, monitoring, incident, and rollback paths.
- Assign corrective work and require a challenge retest.
K: Keep exceptions and continuing review active
An exception requires the unmet requirement, exposure, compensating action, accountable approver, expiry, review date, and escalation trigger. Expired exceptions automatically reopen the queue item.
Continuing review should react to amendments, corrections, guidance, incidents, complaints, audit findings, vendor changes, model or data changes, system expansion, and failed monitoring. At a defined interval, the governing owner reviews late triage, overdue actions, exception aging, test failures, monitoring failures, and records closed without complete evidence.
Worked workflow: AI disclosure across export routes
A fictional organization receives an official change that may require a disclosure in a customer-facing AI assistant. This example does not interpret a real law.
Capture. The monitor stores the official instrument, source version, dates, jurisdiction, and retrieval time. A vendor bulletin is attached only as a discovery aid.
Link. Legal maps the change to the web assistant, PDF exports, support scripts, conversation archive, and a third-party model service. The web and PDF routes are assessed as in scope. Archive treatment remains uncertain and is sent for advice.
Operationalize. Approved requirements cover persistent web and PDF wording, an explanation route, updated support guidance, archive handling, vendor evidence, and post-release sampling. Product, support, records, privacy, and vendor owners receive linked work with due dates.
Confirm. Independent testing finds the disclosure on the web journey but not in PDFs created through a legacy export. The workflow cannot close. Product repairs the renderer, and the tester repeats web, PDF, accessibility, support, and monitoring scenarios against the released version.
Keep. The vendor evidence item remains under a dated exception with restricted exposure. Four weeks of monitoring sample both export routes. Closure states only that the specified requirements passed recorded tests on that version and date. It does not establish legal conformity across the organization.
Queue states and service rules
| State | Entry rule | Exit evidence | | --- | --- | --- | | captured | official source stored | triage disposition | | assessing | material scope question open | approved applicability decision | | implementing | requirement and owner approved | changed artifact and test-ready version | | testing | expected results frozen | result, corrective action, or exception | | monitoring | implementation released | required observation period complete | | closed | critical evidence supported | closure approval and reopen triggers | | held | material uncertainty or failure | authorized resolution or withdrawal |
Failure and escalation conditions
Hold or escalate when the official source cannot be reconstructed, scope reasoning is missing, an effective date may be missed, material requirements lack owners, the implemented version differs from the reviewed version, a critical test fails, a vendor dependency is unresolved, or an exception lacks authority or has expired.
Two Prune can place a participant in a realistic regulatory-change assessment and evaluate how they locate authority, scope effects, use AI, verify evidence, manage exceptions, and communicate a recommendation. It does not provide legal advice, operate the workflow, certify conformity, or make employment decisions.
Sources
FCA: Handbook of rules and guidance
- FCA Handbook FAQ
- EUR-Lex: Regulation (EU) 2024/1689, consolidated text
- NIST AI Risk Management Framework Core
Sources: [4] [2] [3] [1]
Sources
- 1.NIST AI RMF Core · National Institute of Standards and Technology
- 2.FCA Handbook FAQ · handbook.fca.org.uk
- 3.EUR-Lex: Regulation (EU) 2024/1689, consolidated text · eur-lex.europa.eu
- 4.FCA: Handbook of rules and guidance · www.fca.org.uk