MAS confirms AI model risk management guidelines mandatory for Singapore's largest financial institutions by end-2026

The Monetary Authority of Singapore published its consultation response on AI in Finance on April 20, closing an industry engagement process that began with a consultation paper in November 2025.

**The headline decision:** AI Model Risk Management (AI-MRM) guidelines will be mandatory for: - All 5 D-SIBs (DBS, OCBC, UOB, Standard Chartered Singapore, Citibank Singapore) — by 31 December 2026 - Licensed insurers above S$10B assets under management — by 31 December 2026 - All other MAS-licensed financial institutions — by 30 June 2027

The AI-MRM framework requires covered institutions to:

1. **Maintain an AI model inventory** covering all AI systems used in credit decisioning, trading, fraud detection, underwriting, customer service, and regulatory reporting — with risk classification for each 2. **Implement pre-deployment validation** for high-risk models, including independent model validation, bias testing, and explainability documentation 3. **Establish ongoing monitoring** with defined trigger thresholds for model performance degradation and mandatory revalidation schedules 4. **Appoint a Model Risk Officer** (or equivalent) with clear accountability for AI governance

**Industry feedback incorporated:** The consultation response accepts four industry requests: a phased implementation timeline (vs the original simultaneous deadline), exclusion of models below a materiality threshold from full inventory requirements, a recognised third-party validator pathway to reduce internal capacity constraints, and a 12-month safe-harbour period for models already in production at the framework's commencement date.

**AIMenta take:** The MAS framework is well-designed — more operationally grounded than the EU AI Act's risk-tier abstraction, and more specific than the HKMA's principles-based AI guidance. For Singapore-based financial services clients, the 12-month safe-harbour is the critical planning variable: it means models deployed before the framework's effective date are grandfathered, but new model deployments from Q4 2026 onward face immediate compliance requirements. We expect a surge in pre-deadline deployment activity — and a corresponding surge in demand for AI governance documentation support — in Q3 2026.

Regional note: The MAS framework is likely to become the de facto ASEAN reference standard for AI in finance, given Singapore's role as a regional financial centre. Indonesian OJK and Malaysian BNM have both signalled alignment with MAS approaches in prior publications.