Singapore's financial regulator has officially brought AI agents under its risk-management oversight, requiring banks to demonstrate that autonomous systems can be tested, validated, and monitored before they execute real transactions. The Monetary Authority of Singapore (MAS) confirmed in a parliamentary response that its forthcoming AI risk-management guidelines will cover agentic AI, raising the bar for how financial institutions deploy autonomous systems across payments, treasury, wealth management, and other critical workflows.

This marks a significant shift from treating AI agents as experimental tools to treating them as regulated systems that must prove their safety before going live. Unlike traditional AI models that generate recommendations for humans to review, agentic AI can pursue multi-stage objectives, interact with applications, and initiate actions without human approval at every step. That fundamental difference creates new testing challenges that banks will need to solve to satisfy regulators.

What Makes Testing AI Agents Different From Testing Regular AI?

Testing an AI agent is not the same as testing a language model or a recommendation engine. A conventional AI system might be evaluated on accuracy, bias, or response quality. An agent, by contrast, must be tested to ensure it respects its assigned mandates, validates transactions before execution, escalates unexpected behavior, and leaves an auditable trail of every decision.

MAS addressed this challenge directly in July through its Safeguards for Agentic Finance at Runtime framework, known as SAFR, developed with industry partners including HSBC, J.P. Morgan Chase, Mastercard, and Visa. The framework concentrates on governance at the moment an agent proposes an action, embedding "policy bound execution, real time validation, auditability and interoperability" into agentic operations.

For quality assurance and testing teams, this creates a substantial new testing perimeter. The framework requires:

• Policy-Bound Execution Testing: Negative testing to prove that an agent cannot move beyond its authorized purpose, access prohibited systems, or exceed transactional limits.
• Real-Time Validation Testing: Assurance that validation systems work accurately, respond quickly, and handle edge cases like unavailable services or ambiguous results.
• Auditability Testing: Verification that the full decision trail is captured accurately and cannot be altered after the fact.
• Interoperability Testing: Examination of what happens when agents exchange information, delegate work, or interact with legacy applications and external services.

Testing individual agents in isolation will not be enough. A compliant component can still produce an unsafe outcome when combined with another agent, application, or workflow. Banks will therefore need end-to-end scenarios that test interactions, dependencies, and cumulative behavior across complete business processes.

Why Are Financial Institutions Deploying AI Agents Now?

The use cases driving adoption are compelling. MAS envisions "agent-assisted payments and treasury operations, where autonomous agents can execute routine transactions within predefined mandates, improving efficiency and reducing operational frictions." The regulator also points to "wealth management and advisory workflows, where AI agents review documents and generate structured assessments within narrowly scoped task boundaries, supporting faster and more consistent compliance review".

In these environments, an output error can quickly become an operational event. An agent could initiate an incorrect transaction, apply the wrong mandate, use stale information, or trigger a sequence of actions across interconnected systems. That risk is why regulators are now demanding evidence that these systems work safely before deployment.

The regulatory shift reflects a broader industry movement toward operationalizing AI governance. In April, MAS released the MindForge AI Risk Management Toolkit, developed with 24 financial institutions and industry participants. The toolkit is structured around scope and oversight, AI risk management, lifecycle management, and the organizational capabilities needed to support responsible deployment.

"To fully realise AI's value, governance must be treated as a strategic imperative," said Sameer Gupta, Chief Analytics Officer at DBS.

Sameer Gupta, Chief Analytics Officer at DBS

How Should Banks Prepare for Agentic AI Governance?

Financial institutions will need to operationalize AI governance principles through controls that can be tested repeatedly across the AI lifecycle. This includes:

• Pre-Deployment Validation: Testing agents against defined mandates, risk boundaries, and transaction limits before they access live systems or customer accounts.
• Data-Quality Testing: Ensuring that agents receive accurate, timely information and can handle missing or conflicting data without making unsafe decisions.
• Model and Prompt Change Management: Establishing processes to test and approve changes to agent behavior, decision rules, or underlying models before deployment.
• Post-Deployment Monitoring: Continuous observation of agent behavior in production, with alerts for deviations from expected patterns or policy violations.
• Incident Handling and Controls: Procedures for responding to agent failures, escalating decisions to humans when needed, and retiring or replacing agents that no longer meet standards.

Traceability becomes increasingly important as well. Institutions will need to know where AI is operating, which data and third-party components it depends on, what decisions it can influence, and which tests or approvals were completed before deployment.

The regulatory framework is not prescriptive about how banks should achieve these outcomes. MAS confirmed it will maintain a principles-based approach because AI technology and its associated risks are evolving rapidly. However, the parliamentary response makes clear that financial institutions will be responsible for deciding how to satisfy the principles, and they will also need to produce credible evidence that their chosen controls work.

MAS said the Future of Finance Institute will support SAFR through "industry pilots and sandbox experimentation," helping institutions "to test and deploy SAFR-aligned solutions." This suggests that regulators and industry will work together to develop testing methodologies and best practices over the coming months.

For banks, the message is clear: autonomous AI agents are no longer experimental. They are regulated systems that must prove their safety, reliability, and compliance before deployment. The testing burden is substantial, but it reflects a fundamental reality: in financial services, an agent's error is not a recommendation to ignore, but a transaction that has already executed.