MAS

MAS has several core moves on AI governance:

• FEAT principles (2018): Fairness, Ethics, Accountability, Transparency — the four principles setting the minimum bar for AI use by financial institutions
• Veritas framework (2020+): turns FEAT into an executable assessment methodology. Veritas is an open-source AI governance toolkit led by MAS and co-developed with the banking industry (DBS, UOB, OCBC, HSBC, Standard Chartered, etc.)
• Generative AI regulation (2024+): MAS issued specific guidance on financial institutions' use of GenAI — banning the use of GenAI for customer decisions, requiring human oversight, etc.

At the industry level, MAS is not just a regulator but also actively pushes AI deployment:

• AI Talent Push: Talent Programmes that build up AI talent reserves in finance
• Sandbox / TechFin: lets financial institutions test AI products inside the regulatory sandbox
• GAIIN (Global AI Innovation Network): collaborates with financial regulators in Switzerland, the UK, Australia, etc. to build cross-border dialogue on AI governance

MAS's AI governance is a "sector-level + tool-level + international-level" trinity:

• Sector-level: FEAT + Veritas are aimed at financial services and are more concrete than IMDA's MGF (finance has high-stakes AI use cases like credit scoring, anti-fraud, and KYC)
• Tool-level: Veritas provides actual code and assessment methods, not just documents
• International-level: MAS actively pushes Veritas to other countries' financial regulators via GAIIN, building a de facto standard

Technically, the Veritas assessment framework includes:

• Fairness assessment: automated tests for multiple fairness metrics
• Explainability methods: financial-sector adaptations of XAI tools like SHAP and LIME
• Robustness testing: adversarial samples, data drift, and concept drift detection
• Data governance: checklists for training data sources, bias, and privacy

This toolkit is actually deployed at DBS, UOB, HSBC, and others — making it one of the few AI governance tools with real production validation.

MAS's AI governance moves directly define the cadence of AI deployment in Singapore's financial sector.

In the "seven transmission levers" framework:

• Lever 4 (governance): the enforcement body for financial-sector AI governance
• Lever 3 (industry adoption): pushes financial AI deployment via Sandbox + TechFin
• Lever 6 (international): internationalises governance standards via GAIIN

A take: MAS is the institution with the most teeth in Singapore's AI governance system — unlike IMDA's "voluntary adoption + soft standards" route, MAS has real penalty power over financial institutions; FEAT/Veritas is not advice but industry rule.

This also means MAS's AI judgement is existentially important for Singapore's financial sector: too loose, and financial stability is at risk; too tight, and Singapore's appeal as an Asia-Pacific financial hub erodes. MAS's stance on GenAI (requiring human oversight, restricting customer-decision use cases) has skewed conservative — consistent with its tradition of "financial stability over innovation".

Worth watching going forward: when MAS will allow GenAI in direct customer-facing roles (e.g., customer service, robo-advisory), when Veritas will be upgraded for full LLM evaluation, and potential integration with SEA-LION (the financial sector's demand for localised LLMs).