Singapore’s financial services regulator is turning scam detection into a live test of how banks validate artificial intelligence in controlled, compliant and evidence-ready environments.
The Monetary Authority of Singapore’s new proof-of-value project, which brings together data from five banks to explore AI and machine learning for pre-emptive scam detection, should be read as more than a financial-crime initiative.
For QA and software testing teams, it is a clear signal that MAS is accelerating its shift from AI governance frameworks to production-style validation, where models, data environments and control mechanisms must all be tested together.
That shift reflects a broader regulatory direction. MAS is increasingly tying innovation to assurance, requiring banks not just to deploy AI, but to demonstrate that systems behave as intended, that risks are understood, and that controls work in practice rather than on paper.
Multi-bank data
At the centre of the new initiative is a collaborative model training and validation exercise that cuts across institutional boundaries.
MAS clarified it is working with industry partners, the Government Technology Agency of Singapore and the Singapore Police Force to enhance scam detection using AI and machine learning techniques.
The proof-of-value aims to build more robust models by combining datasets from multiple banks, allowing earlier identification of higher-risk transactions and accounts.
“Prompt identification could enable timely assessment, intervention and reduction of customer losses to scams,” MAS said.
For QA teams, that introduces a more complex validation problem than traditional single-institution models.
Combining datasets across banks creates challenges around data consistency, feature alignment, bias, drift and model explainability across different customer bases and transaction behaviours.
Testing therefore moves beyond model accuracy in isolation. It includes validating whether models behave consistently across heterogeneous datasets, whether risk signals remain stable under changing data conditions, and whether detection thresholds produce acceptable levels of false positives and false negatives in live-like environments.
For software testing and QA teams, those controls are not insignifcant or minor. They become part of the system under test.
Validation must extend to whether hashing is implemented correctly, whether access controls prevent unauthorised exposure, whether monitoring systems detect anomalies in data usage, and whether deletion processes execute reliably at the end of the lifecycle.
In effect, the testing perimeter expands from model performance to include privacy engineering, data governance and security controls. The assurance requirement is no longer limited to “does the model work”, but “does the entire data and control environment meet regulatory expectations under real conditions”.
Compliance evidence
MAS’ latest initiative fits directly into MAS’ wider push to operationalise AI governance.
Through its MindForge AI Risk Management Toolkit, developed with major institutions including HSBC, Citi, UBS, BlackRock, DBS, Standard Chartered, UOB and Prudential, MAS has already signalled that responsible AI will be judged by execution.
BlackRock’s Marko Milek said the framework helps translate “responsible AI principles into actionable risk management”, while DBS Chief Analytics Officer Sameer Gupta stressed that “to fully realise AI’s value, governance must be treated as a strategic imperative”.
That emphasis on execution is now visible in the scam-detection project. Rather than setting expectations at a policy level, MAS is effectively creating a supervised environment in which banks must demonstrate how governance works in practice.
For QA teams, that means generating evidence across the AI lifecycle: pre-deployment validation, data integrity checks, model testing, monitoring thresholds, alerting behaviour, escalation workflows and post-deployment performance tracking.
The result is a shift from documentation-heavy governance to evidence-based assurance, where testing outputs, logs, metrics and audit trails form the backbone of compliance.
Financial-crime control
The choice of scam detection as a use case is significant. Financial crime is one of the most operationally sensitive domains in banking, where detection speed, accuracy and explainability directly affect customer outcomes and regulatory exposure.
AI models deployed in this space must perform under real-time conditions, handle high data volumes and integrate with complex operational workflows.
For QA and software testing teams, that brings them closer to the core of financial-crime control than before.
Testing responsibilities extend into validating alert logic, ensuring that flagged transactions are correctly routed, verifying that intervention processes trigger as expected, and confirming that human review workflows can handle model outputs at scale.
The challenge is not simply technical. It is regulatory. Banks must be able to show supervisors that their detection systems are reliable, that controls are functioning, and that outcomes can be explained and audited.
That aligns with MAS’ broader stance that AI systems in finance must be observable, testable and governable across their lifecycle.
Wider MAS strategy around testing
The project also reflects MAS’ longer-term strategy of embedding testing, resilience and collaboration into financial technology adoption.
The regulator has previously called for stronger oversight of third-party and open-source software, warning that “financial institutions can no longer treat third-party risks as peripheral”.
It has also pushed banks to maintain detailed inventories of technology dependencies, a requirement that becomes even more critical when AI systems rely on external data sources, APIs and infrastructure components.
In parallel, MAS has supported industry-wide testing initiatives, from quantum security trials to cross-border AI validation efforts with the UK’s Financial Conduct Authority.
Jessica Rusu, the FCA’s Chief Data, Information and Intelligence Officer, said that partnership would “be championing safe and responsible AI innovation across UK and Singapore markets”.
Across these initiatives, the pattern is consistent. MAS is not relying on firms to self-certify AI systems. It is creating shared environments, collaborative projects and structured frameworks where testing, validation and monitoring can be carried out under realistic conditions.
Testable infrastructure
The scam-detection proof-of-value highlights another key element of MAS’ approach: treating AI as shared financial infrastructure rather than isolated innovation.
MAS has previously pointed to “strong prospects for the financial industry to apply AI to solve industry-wide problems beyond what each financial institution can do individually”, while also arguing that Singapore can become a “centre of excellence” for AI testing and deployment in financial services.
The new project brings those ideas together. By combining data across banks within a controlled environment, MAS is exploring whether industry-level AI models can deliver better outcomes while maintaining strict privacy and security standards.
For QA teams, that introduces a new layer of complexity. Industry-wide systems require consistent testing standards, shared validation methodologies and aligned control frameworks. They also raise questions about accountability, auditability and cross-institutional coordination.
Ensuring that such systems meet regulatory expectations will depend heavily on the ability of testing teams to produce reliable, comparable and reproducible evidence.
Foundation of trust
MAS said the current proof-of-value “lays the groundwork for deeper industry collaboration” and may be expanded to include broader datasets and more sophisticated models.
If that happens, the testing and assurance requirements will scale accordingly. Larger datasets, more complex models and additional use cases will increase the need for continuous testing, real-time monitoring and robust control validation.
For banks, the direction is becoming increasingly clear. AI adoption will be judged not by the sophistication of models alone, but by the strength of the testing, governance and compliance frameworks that support them.
For QA and software testing teams, that cements their role at the centre of AI strategy.
MAS is not just encouraging banks to use AI. It is requiring them to prove that it works safely, securely and consistently in production-like conditions.
In doing so, it is turning financial-crime detection into a test case for the future of AI assurance in banking, where testing is no longer a supporting function, but the foundation of trust.
THIS MONTH
Why not become a QA Financial subscriber?
It’s entirely FREE
* Receive our weekly newsletter every Wednesday * Get priority invitations to our Forum events *
REGULATION & COMPLIANCE
Looking for more news on regulations and compliance requirements driving developments in software quality engineering at financial firms? Visit our dedicated Regulation & Compliance page here.
READ MORE
- Trust, not speed: Why AI governance is now a testing battleground for banks
- NatWest’s AI trade finance overhaul opens new chapter for QA teams
- Banking UAT moves beyond sign-off as QA takes centre stage in system rollouts
- Citi ramps up AI-driven testing in race to modernise legacy systems
- Lloyds, HSBC and NatWest get OpenAI access amid mounting concerns
WATCH NOW
QA FINANCIAL PODCASTS
CLICK HERE TO LISTEN TO OUR EXCLUSIVE CONVERSATIONS
