Summary
- Issued 7 Oct 2026 after a Nov 2025 consultation. They take effect 7 Oct 2027, with life cycle controls due by 7 Oct 2028 (para 1.8).
- Apply to all FIs and all AI, including generative AI and AI agents, proportionate to size and risk (paras 1.5, 2.1).
- Low-impact shortcut: if AI failure would not materially hurt the FI or customers (e.g. drafting emails, internal summaries), a basic policy set is enough: a named senior owner, allowed and banned uses, an approved tool list, staff education, compliance checks (paras 2.3 to 2.5).
- Core build (Sections 3 and 4): board and senior management oversight, AI in the risk appetite, a control function that identifies AI use (including AI hidden inside SaaS), an AI inventory, and a risk materiality rating based on impact, complexity and reliance.
- Life cycle controls (Section 5): data management, transparency and explainability, fairness, human oversight, third-party AI, model selection, testing, cybersecurity, reproducibility, independent validation for high-risk uses, monitoring with kill switches, change management, decommissioning.
- Earlier layers still apply: the 2018 FEAT principles (Fairness, Ethics, Accountability, Transparency) continue to guide AI use (para 1.2). The Guidelines also point FIs to IMDA's Model AI Governance Framework, the AI Verify Foundation initiatives, and IMDA's agentic AI framework (footnotes 3 and 11).
Who it applies to
- All financial institutions as defined in the Financial Services and Markets Act 2022, on a group basis for some locally incorporated FIs (paras 1.1, 1.2).
- Not directly to vendors. But FIs keep full accountability for third-party AI and must push requirements into contracts (paras 5.10, 5.11), so vendors feel it through procurement.
Key dates
- MAS consults on proposed Guidelines
- Consultation closes
- Final Guidelines issued with response to feedback
- Guidelines take effect; Sections 3 and 4 (oversight, identification, inventory, risk materiality) expected from this dateUpcoming
- Sections 5 and 6 (life cycle controls, capability and capacity) expected by this dateUpcoming
Full date record
| Date | Event |
|---|---|
| 2018 | FEAT principles published by MAS |
| 2025-11-13 | Consultation paper issued |
| 2026-01-31 | Consultation closes |
| 2026-10-07 | Final Guidelines issued |
| 2027-10-07 | Effective; Sections 3 and 4 expected |
| 2028-10-07 | Sections 5 and 6 expected |
Penalties and enforcement
- No fines in the Guidelines. MAS calls them supervisory expectations; it assesses FIs against them in inspections.
- Local senior management of global groups must be able to show MAS how they oversee AI, even if they use group frameworks (para 3.6).
What it means for an Indian company
- Indian IT and fintech vendors serving Singapore banks and insurers should expect contract clauses on: notice before adding AI to a service, notice of model updates, right to audit, data protection and performance guarantees (para 5.11(f)).
- Prepare a vendor AI pack: model documentation SC-AI-014 Model documentation (model cards, system cards), test results SC-AI-015 Pre-deployment testing and validation, fairness results for credit or underwriting tools SC-AI-017 Fairness and discrimination testing of outputs, red team results SC-AI-018 Robustness and adversarial testing (red teaming).
- Independent certifications or external assessments carry more weight than self-attestation (para 5.11(a)); ISO/IEC 42001 certification helps here (SC-AI-039 Conformity assessment and certification readiness).
- Disclose AI embedded in your SaaS features; FIs must identify it (SC-AI-003 AI system inventory, SC-AI-007 Third-party and vendor AI due diligence).
- Support change notifications and rollback (SC-AI-019 Change management and approval gates), monitoring feeds (SC-AI-020 Performance monitoring and drift detection) and kill switches or fallbacks (SC-AI-033 Business continuity and fallback for AI).
- Very close in shape to RBI FREE-AI, so one evidence set can serve Indian and Singapore bank clients.
Obligations
Full obligation breakdown coming
We break the most requested policies down into numbered obligations mapped to controls. Leave your details and we’ll tell you when MAS AI Risk Management Guidelines is done.
Glossary
- FI: financial institution regulated by MAS.
- Risk materiality: how much harm an AI use case could cause, rated by impact, complexity and reliance.
- AI inventory: register of AI use cases, systems and models with owners and risk ratings.
- Kill switch: a way to switch AI off quickly if it goes beyond risk limits.
- FEAT: MAS 2018 principles of Fairness, Ethics, Accountability and Transparency for AI and data analytics.
- AI Verify: IMDA's AI governance testing framework and toolkit, now run through the AI Verify Foundation.
Key takeaways
- Final, supervisory, sector-wide AI rulebook for Singapore finance; effective 7 Oct 2027, full life cycle controls by 7 Oct 2028.
- Proportionate: a light "basic policies" track for low-impact AI, full controls for material use cases.
- Vendors are pulled in through third-party AI clauses and testing expectations.
Official sources
- Guidelines page, MAS, accessed 2026-10-19
- Guidelines PDF (7 Oct 2026), accessed 2026-10-19
- Media release, 7 Oct 2026, accessed 2026-10-19
- Consultation media release, 13 Nov 2025, accessed 2026-10-19
- FEAT principles (2018), accessed 2026-10-19
This is general information, not legal advice. Last verified 19 Oct 2026 against the official source. Laws change; confirm the current position with qualified counsel before acting.