MAS Proposes AI Risk Guidelines for Banks, Holding Boards and Senior Management Accountable

MAS sets out AI risk management guidelines: Boards are accountable

Singapore's central bank has published new AI Risk Management Guidelines that put bank boards and senior leaders on the hook for managing AI risks. The guidance lays out expectations for governance, systems and policies, lifecycle controls, and the skills needed to deploy AI responsibly across the business.

The Guidelines apply to a wide range of AI tools and use cases, including Generative AI and newer setups like AI agents. MAS is seeking industry feedback, with comments due by 31 January 2026.

What MAS expects from leadership

• Board and senior management must oversee AI risk management and set the tone for risk culture.
• Establish clear frameworks, structures, policies, and processes for AI governance across the enterprise.
• Maintain an accurate, up-to-date inventory of all AI systems and use cases.
• Run risk materiality assessments that consider impact, complexity, and reliance.

Controls across the AI lifecycle

Banks are expected to put effective controls in place at each stage of the AI lifecycle. Key areas include:

• Data management and quality
• Fairness and bias mitigation
• Transparency and explainability
• Human oversight and accountability
• Third-party and model vendor risk
• Evaluation, testing, and validation
• Monitoring and change management

Ho Hern Shin, MAS Deputy Managing Director, said the guidelines set clear supervisory expectations and support responsible innovation for institutions that put the right safeguards in place.

What to do now (for boards and executives)

• Assign a single accountable executive for AI risk with a direct line to the board risk committee.
• Stand up an enterprise AI inventory; classify use cases by materiality and reliance on third parties.
• Update your risk appetite statement to include AI risk, with thresholds for explainability, bias, and model drift.
• Integrate AI models into existing model risk management and ICT/operational risk processes.
• Tighten vendor due diligence for foundation models, APIs, and AI agents; define exit and fallback plans.
• Document data lineage and feature pipelines; require human-in-the-loop for material decisions.
• Set continuous monitoring and revalidation schedules; log decisions and changes for auditability.
• Run targeted training for product, risk, audit, and front-line teams on AI controls and escalation.

Timeline and next steps

MAS has opened a consultation and invites comments until 31 January 2026. For context on Singapore's approach to responsible AI, see MAS's resources on responsible AI policy direction here.

Upskill your teams

If you're setting up board-level oversight and first-line controls, give your managers and risk teams practical training on AI use in finance. Explore role-based learning paths at Complete AI Training - Courses by Job.