MAS third-party AI: build an assurance register

MAS third-party AI expectations keep accountability with the financial institution even when a model or service is built, operated or supplied by another party. A practical assurance register links each external AI dependency to its intended use, risk materiality, evidence, unresolved gaps and decision owner. It should support action, not become another vendor list.

Map the service to the delivered financial activity

Record the provider, model or service, business owner, users, data, integrations and decisions it influences. Include embedded AI in software, intragroup services and subcontracted components where known. Describe the customer or operating effect if output is wrong, unavailable, manipulated or disclosed. This context determines how much assurance is proportionate.

The MAS Guidelines on Artificial Intelligence Risk Management apply to all financial institutions and all forms of AI, with effect from 7 October 2027. The accompanying MAS release explains that firms remain accountable for AI used in their services, including AI developed, operated or provided by third parties.

Request evidence that answers the real risk

Match evidence requests to materiality and the AI life cycle. Relevant items may include model purpose, testing, data controls, security, human oversight, monitoring, incident handling, change notices, subcontractors and business continuity. A general certificate or policy can inform review, but it does not show that the service is suitable for the institution’s specific use.

Record the evidence date, scope, source, reviewer and limitations. Distinguish documents received from claims verified. Where providers cannot disclose details, identify the resulting assurance gap and consider practical alternatives such as constrained use, additional testing, output review, data minimisation, monitoring or contractual notification rights.

Make assurance gaps lead to a decision

Assign each gap a severity, action, owner and expiry date. Compare residual risk with approved risk appetite. The MAS release says institutions should consider limiting, suspending or replacing a third-party AI service when risks cannot be brought within appetite. Record that choice and the operational path, including migration, customer impact and record retention.

Link the register to the MAS AI readiness map and the AI vendor data review. The business support cornerstone can help separate implementation funding from the institution’s continuing accountability.

Keep one controlled record

Create a dated record that connects the decision, current official source, approved scope, responsible owner, milestones, costs and evidence. Distinguish proposed, approved, completed, submitted and paid items. A portal receipt, report or invoice proves one event, not the complete outcome. Preserve earlier versions and explain material changes instead of replacing history.

Assign each document to a named owner and use consistent identifiers across quotations, contracts, invoices, payments and deliverables. Where an authority or professional adviser must decide a point, record the dependency instead of substituting an internal assumption. Management remains responsible for declarations, commercial choices and continuing obligations.

Review the work in stages

Use four reviews: suitability, authority, implementation and closeout. At suitability, confirm the entity, activity and business need. Before commitment, freeze the proposed scope and supporting facts. During implementation, compare actual work with approved terms. At closeout, reconcile deliverables, cost, payment, operating results and obligations that continue.

Name the person who may approve each review and the evidence that person must see. Procurement should not treat a draft application as authority to contract. A project team should not treat vendor completion as company acceptance. Finance should not treat claim submission as receipt. These distinctions protect both operating quality and the integrity of declarations.

Manage exceptions without rewriting history

Maintain an exception log for delayed milestones, changed suppliers, cost movements, technical limits, missing evidence and failed tests. State the effect on scope, eligibility, timing and business value. Where authority approval is needed, use the published route before relying on the change. Silence, a draft upload or a helpdesk acknowledgement is not approval.

If a material defect appears, pause the affected acceptance, filing or claim item. Preserve original evidence and assign corrective action. Do not backdate documents, move costs between categories or alter a baseline after seeing the result. A transparent exception with an owned resolution is more credible than a clean file that cannot explain the actual project.

Finish with independent readback

Ask a person who did not prepare the final pack to compare the decision, current official source, approved terms, deliverables, invoices and operating evidence. Recalculate totals and test source records. Record unresolved items with owners and dates. The reviewer should be able to reproduce the conclusion without relying on the preparer’s memory.

After submission, preserve the receipt and reconcile questions, adjustments and final outcomes to the controlled record. Continue retention, reporting, asset-use or other conditions that survive closeout. Completion means management can explain what was decided, what was delivered, what supports it and what still needs attention.

Schedule a later review suited to the decision. Compare expected results with actual operating, financial and compliance evidence. Record benefits that did not appear, new risks and actions still open. Before a later commitment or filing, check the live official source again without deleting the evidence used for the earlier decision.

Keep the decision usable for the people who operate it. Provide a short handover that identifies the current version, responsible owner, next review, open exceptions and source locations. Restrict sensitive records to authorised users, while giving each operational owner enough information to perform the assigned control. Test the handover with a real item rather than assuming that a shared folder is understood.

Practical review checklist

• Third-party AI uses inventoried
• Materiality assessed
• Evidence scope recorded
• Assurance gaps owned
• Limit, suspend or replace decision documented

A MAS third-party AI register is valuable when missing assurance changes the service decision. The institution should be able to show what it knows, what it could not verify, which controls compensate and who accepted the remaining risk.