On 7 October 2026 the Monetary Authority of Singapore issued Guidelines on Artificial Intelligence Risk Management for the financial sector. They apply to all financial institutions and all forms of AI, including more autonomous systems. Most of the commentary has focused on what firms must do. The more interesting question is what the guidelines ask of supervisors.
What MAS expects of firms
The expectations are clear and proportionate. Boards and senior management should oversee AI risk. Firms should keep an inventory of where they use AI, assess how material each use is, and apply controls across the AI lifecycle: data governance, testing, human oversight, cybersecurity, monitoring and change management. And firms stay accountable for AI that third parties develop, operate or provide.
What that asks of supervisors
Every one of those expectations creates supervisory work. Someone has to review the inventories, test whether materiality assessments are sensible, and check that human oversight is real rather than nominal. Multiply that across every supervised firm and it becomes a data problem: structured information, collected consistently, that can be compared across a population.
The bar applies to supervisors too
Supervisors are adopting AI fast: three in four now use it, according to Central Banking's 2026 benchmarks. Yet the BIS found only 4% of central banks rate their ability to monitor AI governance as high. A regulator asking firms to inventory, assess and oversee their AI will be expected to show the same discipline with its own. Predictable AI, where deterministic rules make the calls that must be repeatable and every finding traces to its source, is how supervisors meet that bar.
What supervisors elsewhere can take
Singapore's approach is principles-based and proportionate, which makes it portable. Authorities that have not yet set AI expectations for firms can borrow its structure. And all authorities can apply it to themselves first.
- 1Keep an inventory of every place your authority uses AI, including ad hoc tools.
- 2Rate each use by materiality: what happens to a supervisory decision if it is wrong?
- 3For material uses, make deterministic rules take the decision and let AI do the reading.
- 4Require every AI-assisted finding to be traceable to its data and rule.
- 5Hold your vendors to the same accountability you expect firms to hold theirs.
Source: MAS, media release, 7 October 2026. For where your authority stands today, take thethree-minute readiness check, or read what SupTech is.