Singapore MAS Guidelines on Artificial Intelligence Risk Management (2026) 2.3-2.5: Low-impact AI use may rely on basic AI governance: assess potential impact on the FI and on customers, and keep at minimum senior management accountability, permitted and prohibited uses with human oversight, an approved tool list and approval process, staff education, compliance checks and periodic review of continued eligibility

What does 2.3-2.5 require?

A lighter regime exists for AI whose failure would not matter much. An FI may run basic AI governance where poor performance or unavailability of the AI tools it uses is unlikely to harm the FI, its customers or other stakeholders materially; the test weighs financial, operational, regulatory, legal and reputational consequences for the FI and the consequences for customers and stakeholders, fairness, ethics and consumer protection included. MAS gives examples that would usually pass: drafting or polishing customer emails, summarising documents or meeting notes internally, first-pass document review, producing formulas, charts or visuals for internal use, generating images for marketing or internal material, and chatbots that point staff to internal policies, always with a human checking the output before it is used. The basic regime still needs, at least: a named senior manager accountable for AI oversight; rules on what AI may and may not be used for (such as a ban on putting confidential, proprietary or client information into public tools) backed by real human oversight, including when a person must check and validate outputs before use; a list of approved AI tools and a route for approving new ones; staff education on the AI rules; regular compliance checks; and a review, both periodic and event-driven, of whether the FI's AI use still qualifies as low impact.