MAS and FCA Launch Joint Partnership for Responsible AI in Finance
AI is moving from pilots to production. To keep pace without tripping over fragmented rules, the Monetary Authority of Singapore (MAS) and the UK's Financial Conduct Authority (FCA) have launched the UK-Singapore AI-in-Finance Partnership.
Announced at the Singapore FinTech Festival, this is a practical step to help fintech providers and financial institutions scale AI safely across two major financial hubs. For teams in the UK and US, the message is clear: trustworthy AI is becoming a cross-border standard, not a nice-to-have.
Why this matters now
AI in finance is shifting from individual tools to connected, agentic systems. That raises the stakes on model risk, bias, explainability, and operational resilience.
Without common expectations, firms pay twice-once to build, and again to retrofit for each market. This partnership aims to cut that drag by aligning what "safe and scalable" looks like.
What's actually in the partnership
- Joint testing and regulatory insights: MAS and FCA will co-test AI solutions and share supervisory feedback. Firms can validate against converging standards before wide release, reducing time-to-market and rework.
- Sandboxes and pipelines: The work builds on MAS's PathFin.ai and the FCA's AI Spotlight, creating channels to share quality solutions and research across both markets.
- FCA presence in Singapore: The FCA is establishing a formal presence via a Financial Services Attaché at the British High Commission to deepen day-to-day coordination with MAS.
For reference, see the regulators' homepages: MAS and FCA.
What supervisors will expect
- Model risk governance and accountability: Clear ownership, board oversight, and decision rights across the model lifecycle.
- Bias detection and data quality: Documented datasets, fairness testing, re-sampling strategies, and ongoing monitoring.
- Explainability standards: Especially for customer-facing tools and high-impact decisions like credit and fraud.
- Operational resilience: Stress testing, fallback modes, human-in-the-loop controls, and incident response playbooks.
MAS has also proposed new Guidelines on AI Risk Management for FIs, emphasizing board and senior management accountability, comprehensive AI inventories, and controls over data management, fairness, and transparency. Expect those themes to show up in joint testing and supervisory dialogue.
Where this moves the needle first
- Credit underwriting: Feature governance, bias controls, challenger models, and clear adverse-action reasoning.
- Compliance monitoring: Model-driven surveillance for AML, market abuse, and conduct with traceable alerts and measurable effectiveness.
- Operational automation: Agentic workflows for customer service and internal processes, with audit trails and rollback plans.
What to do now (practical checklist)
- Stand up AI governance: Assign accountable executives, set risk appetite, and align approval gates with model impact tiers.
- Build an AI inventory: Catalogue models, datasets, providers, prompts, fine-tunes, and downstream dependencies. Track purpose, risk rating, and owners.
- Ship with explainability: Choose model classes and techniques that support reason codes and customer-level explanations where required.
- Operational readiness: Define kill-switches, fallback policies, escalation paths, and recovery time objectives. Test them.
- Bias and data controls: Establish pre-deployment and ongoing fairness tests, data lineage, and drift detection. Document thresholds and actions.
- Third-party risk: Bake audit rights, incident reporting, and model transparency into vendor contracts. Validate claims with evidence, not slides.
- Cross-border by default: Map controls to both MAS and FCA expectations so one build works in multiple markets. Use sandbox opportunities early.
- Training and accountability: Upskill product, risk, compliance, and engineering teams on model risk and supervisory expectations.
Why this benefits your roadmap
Common standards lower rework, speed approvals, and reduce operational surprises. That means cleaner business cases for AI in lending, payments, wealth, and compliance-without a spike in legal or audit costs later.
If you're mapping capabilities and tools for finance use cases, this curated list can help as a starting point: AI tools for finance.
The bottom line
The MAS-FCA partnership is a signal: trustworthy AI in finance is becoming a shared standard, tested in the open, and built to scale. Firms that align governance, testing, and documentation now will move faster with fewer surprises-and be ready for cross-border deployment when the door opens wider.
Your membership also unlocks:
