The Monetary Authority of Singapore (MAS) issued its final Guidelines on Artificial Intelligence Risk Management on 7 October 2026. The first deadline is 7 October 2027: firms need senior management oversight, visibility into where AI is being used, an assessment of how material each use is, and clear accountability. Lifecycle controls and the capability to operate them follow by October 2028. The dates matter. But the bigger issue is whether firms can demonstrate that AI is actually under control.
These are supervisory guidelines rather than legislation. That does not make them optional from a risk-management perspective. MAS is setting out what its supervisors expect to see across regulated financial institutions and across AI technologies, from traditional machine learning to autonomous agents.
You cannot manage AI risk if you do not know where AI is operating. MAS expects board and senior management accountability, AI risk reflected in the firm’s risk appetite, and an inventory that captures actual use cases. Each use case needs to be assessed for impact, complexity and reliance. From there, firms can apply appropriate controls for data, testing, human oversight, cybersecurity, monitoring and change management.
The answer is not necessarily another committee. If existing governance structures can assign ownership, challenge decisions and produce evidence of oversight, use them. The test is whether the controls work.
Some of the highest-risk AI activity is the activity nobody formally introduced. AI can be built into a vendor platform, added through a software update or used by an employee through a public tool. MAS explicitly addresses embedded vendor AI and shadow AI. It also recognizes that no firm will discover every instance. That is why acceptable-use rules, data loss prevention and network monitoring matter: they reduce exposure even when visibility is incomplete.
Third-party AI also needs a practical, risk-based assessment. A vendor using AI to improve an internal workflow is not the same as a system generating outputs that influence portfolio or client decisions. Your firm remains accountable for the service it relies on. The more consequential the AI output, the more assurance you should require. If the risk is within appetite, document the decision. If it is not, add controls or reconsider the arrangement.
This is where proportionality matters. MAS allows basic governance for AI whose failure would not materially affect the firm or its customers. Examples include drafting emails, summarizing documents, preliminary document reviews and internal policy chatbots. But basic governance is not no governance. It still calls for an accountable senior manager, permitted-use rules, an approved tools list, training and periodic reassessment.
For investment managers, the distinction can change quickly. MAS identifies fund management as an area where transparency and explainability are particularly important. A tool that summarizes research is one thing. A tool whose output begins to influence an investment decision is another. Controls need to follow the use case, not the label on the software.
Two details should not get lost. A Singapore entity can use a group-wide framework, but local senior management still owns its obligations and must be able to demonstrate compliance. For AI agents, firms should document access, monitor activity and test their ability to stop an agent when necessary. MAS has also indicated that further guidance on agentic AI is planned for 2027.
“The real issue is not whether a firm has an AI policy. It is whether the firm knows where AI is being used, what data it can access, who is accountable and what happens when something goes wrong. AI can deliver real value in research, document review, DDQs and market analysis. We should be enabling those uses, not blocking them by default. But the security fundamentals do not change because the technology is new. Identify the exposure, restrict access, monitor activity and be able to demonstrate your controls. Build on the security programme you already have.”
Chad Fullerton, VP of Information Security, ECI
Singapore is putting a clear timetable around questions other regulators are already asking. The Financial Stability Board is consulting on sound practices for AI adoption. In Europe, DORA addresses ICT third-party risk, while relevant high-risk obligations under the EU AI Act are scheduled to apply from December 2027. In the UK, the FCA has so far favored applying existing frameworks, including the Consumer Duty and the Senior Managers and Certification Regime, rather than creating a separate AI rulebook.
In the US, firms are also being assessed against existing obligations. FINRA’s 2026 oversight report includes a dedicated section on generative AI. The SEC has focused examinations on AI-related policies. NYDFS has published cybersecurity guidance addressing AI, including a May 2026 advisory on frontier models. A Treasury-supported industry partnership also published a Financial Services AI Risk Management Framework in February 2026, drawing on the NIST AI RMF.
The regulatory language differs, but the operational questions are familiar. Where is AI in use? Who owns it? What information can it reach? How much does the business depend on its output? What are your vendors doing? And can you show the evidence? Firms should be ready to answer those questions using controls they can map across MAS, NIST AI RMF and ISO/IEC 42001, rather than building a separate programme for every jurisdiction. As ECI Chief Innovation Officer Rich Itri has argued in Full Throttle, Full Control and The Future Is Now, much of the regulatory foundation already exists.
AI risk should not sit in a silo. At ECI, we approach it through the same connected disciplines that protect the rest of a firm’s technology environment: access management, data protection, monitoring, vendor oversight and compliance. ECI’s ELLA platform supports the use of leading AI models in a governed environment, with data protection and visibility into usage. Our security and AI teams can help firms assess where they stand and put appropriate controls in place.
Our forthcoming paper on continuous threat exposure management will explore how AI exposure can be considered alongside vulnerabilities, cloud, SaaS and compliance risk, rather than treated as an isolated category.
If you want to understand what the MAS Guidelines mean for your firm’s AI use and security controls, speak with your ECI account team or visit eci.com/ella.
