Singapore’s MAS Has Finalised AI Guidelines For Finance – Is It A Good Approach For Other Countries?

Writing rules for AI in finance is a bit like drafting match rules for a match that’s already kicked off.

In early October, the Monetary Authority of Singapore (MAS) finalised its attempt, a set of AI risk management guidelines for financial institutions that take effect on 7 October 2027.

Under the guidelines, boards and senior management must oversee AI risk. Financial firms need to keep an inventory of where they use AI, followed by a risk assessment of each use so controls match the risk.

Businesses also remain accountable for AI supplied by vendors. But where an AI failure would be unlikely to materially affect the firm, its customers or other institutions, basic policies and procedures are enough. MAS Deputy Managing Director Ho Hern Shin said the aim is for firms to “innovate with confidence, while maintaining the trust of customers”.

The timeline is staggered. The guidelines take effect on 7 October 2027, with some sections phased in by 7 October 2028, and MAS plans to consult the sector in 2027 on whether further guidance on agentic AI would be useful.

It lands just ahead of the Singapore FinTech Festival, which runs from 18 to 20 November at Singapore EXPO and brings the city’s fintech community together.

Tight Controls For Risky AI, Basic Policies For The Rest

The guideline setup comes down to proportionality.

Institutions don’t need to set up a dedicated AI committee, because existing leadership teams can manage the workload as long as their oversight holds up. Controls grow with risk across every stage of the AI lifecycle, covering data governance, testing, human oversight, cybersecurity, ongoing monitoring and change management, and businesses should be reviewed regularly.

The same applies to AI bought from outside vendors. Institutions must get enough assurance that a vendor’s AI suits its intended use, apply compensating controls where the assurance falls short and, if the risks fall outside the institution’s own risk appetite, consider limiting, suspending or replacing the service. Buying in AI doesn’t hand the responsibility over to the vendor, because the firm stays accountable for it.

We asked experts: can other countries follow Singapore’s approach to regulating AI in finance?

Our Experts

• Alex Goryachev, Innovator-in-Residence at Tulane University
• Michael Heinrich, Co-Founder and CEO of 0G Labs
• Nikki Stone, Founder of YQ Advisory
• Edward Tian, Co-Founder at GPTZero

Alex Goryachev, Innovator-in-Residence, Tulane University

“In my work on AI standards with ISO, IEEE and ANSI, and on AI policy with national governments and NGOs, I see Singapore’s approach as the model other countries could follow. I worked in Singapore a lot, including building Cisco’s Global Innovation Center in Singapore and saw firsthand how the country turns ambition into practical rules.

“Regulating by materiality rather than technology keeps rules relevant as models change. The one gap is speed: a risk rating set at launch needs regular refreshes.

“Vendor accountability is workable for smaller fintechs if they share vendor assurance instead of each auditing alone.

“For agents that act on their own, regulators should require a named owner, scoped permissions and a fast way to stop them. MAS is right to start that work now.

“Singapore’s compact market makes close supervision easier, but the AI inventory, board accountability and proportionate controls can work anywhere and I hope they will.”

Michael Heinrich, Co-Founder and CEO, 0G Labs

“Singapore’s relatively concentrated financial sector and close coordination between regulators and industry certainly make implementation easier, but the underlying principles of proportionate oversight, institutional accountability and technical transparency are applicable almost everywhere. I’d particularly like to see emerging financial markets adopt similar frameworks, because AI could substantially reduce the cost of delivering financial services in regions where traditional banking infrastructure remains expensive or inaccessible.

“The next challenge is international interoperability. AI agents will operate across jurisdictions, financial networks and digital asset ecosystems, and we need common technical standards for identity, verification and transaction authorisation to support that activity.”

Nikki Stone, Founder, YQ Advisory

“MAS’s materiality-based approach is the right call: regulate by risk, not by technology, so the rules keep pace with agentic AI. But materiality only holds if accountability travels with the AI, not just the institution. When a vendor’s model makes the call, the bank still owns the outcome, so vendor contracts need the same human-in-the-loop checkpoints MAS asks of internal systems.

“It should travel well past Singapore too: any regulator chasing agentic AI ends up asking the same question. Who’s allowed to pull the plug, and when?”

Edward Tian, Co-Founder, GPTZero

“I’m on the vendor side of this since we sell AI to schools and universities. [Vendor accountability] works if the vendor can show something the firm can check. Schools stopped taking our accuracy numbers at face value, so they test the tool on their own student writing before they sign.

“That’s assurance a small fintech can get. It tells them more than a vendor’s own report. I’d also ask for results by group of customers because our overall score hid a problem with non-native English writers until outside researchers found it.

“I’d rate each use by how easy a mistake is to undo and how fast someone would catch it. Start with the agent suggesting and a person deciding.”