The Monetary Authority of Singapore has launched a Proof-of-Value exercise using artificial intelligence and machine learning to detect scams before they result in customer losses, drawing on pooled transaction data from five banks. The initiative, announced on 4 May 2026, marks the first time Singapore’s central bank has coordinated a multi-institution AI model training exercise specifically for financial crime prevention.
Key Facts At A Glance
- MAS announced the Proof-of-Value on 4 May 2026, in collaboration with the Government Technology Agency of Singapore and the Singapore Police Force.
- Historical transaction data from five participating banks is being used to train and evaluate the AI and machine learning models.
- Bank account numbers in the dataset are protected through hashing, a one-way process that prevents any participating bank from identifying data originating from another institution.
- Data access is restricted to authorised personnel in a continuously monitored, controlled environment.
- All data used in the exercise will be deleted upon its conclusion.
- MAS has indicated it may expand the scope to broader datasets and additional use cases if the Proof-of-Value demonstrates effectiveness.
- The exercise complements Singapore’s existing anti-scam measures, including the PayNow nickname removal announced by ABS on 29 April 2026 and the Shared Responsibility Framework governing banks, telcos, and platforms.
Singapore’s financial regulator is taking a data-pooling approach to AI-driven scam detection, a model that individual banks have so far been unable to replicate on their own.
The Problem With Single-Institution Detection
Financial crime prevention in Singapore has historically been handled at the individual bank level, where each institution trains its own fraud detection models using only the transaction data it holds. The limitation is structural. Scam networks operate across institutions, moving funds between accounts held at different banks in rapid succession. A transaction that appears unremarkable when viewed inside one bank’s dataset may be clearly anomalous when placed alongside data from a second or third institution.
MAS’s Proof-of-Value is designed to address this gap directly. By aggregating historical transaction data, including bank account numbers, from five participating banks, the exercise aims to build AI and machine learning models that identify higher-risk transactions and accounts with greater accuracy than any single-institution system can achieve. The names of the five participating banks have not been disclosed in publicly available materials.
Privacy Architecture And Governance
The privacy dimension of a multi-bank data-sharing exercise of this nature is significant. MAS addressed this through the use of hashing, a one-way algorithmic process that substitutes actual account numbers with a unique set of generated values, ensuring that each bank can only identify its own data within the shared environment. No bank in the consortium can reverse-engineer or access account-level records belonging to another institution.
Access to the pooled data is confined to authorised personnel working within a controlled setting that will be monitored continuously throughout the exercise. All data will be deleted at the end of the Proof-of-Value. The governance structure reflects MAS’s broader approach to data stewardship under Singapore’s Personal Data Protection Act framework and is consistent with principles the regulator has previously articulated in its AI Model Risk Management guidance published in 2024.
Scope And Potential Expansion
MAS described the current exercise as groundwork for deeper industry collaboration rather than a final deployed system. If the Proof-of-Value demonstrates that pooled, multi-institution AI models can improve scam detection outcomes, MAS has indicated it may expand the scope to incorporate broader datasets, a wider set of financial institutions, and a more diverse range of use cases beyond the current transaction-level analysis.
The exercise sits within a wider set of anti-scam measures that Singapore has accelerated in 2025 and 2026. Scam-related losses in Singapore totalled S$913 million in 2025. In 2026, phishing campaigns impersonating major banks have already generated documented losses. The Shared Responsibility Framework, which came into force in 2024, distributes liability for certain scam losses among banks, telcos, and digital platforms. The PayNow nickname removal announced by the Association of Banks in Singapore on 29 April 2026, effective 6 June 2026, closes a related social-engineering loophole.
Broader Implications For ASEAN
The governance model MAS is piloting, where a central regulator coordinates cross-institution AI training on privacy-preserved transaction data, has no direct precedent in Southeast Asia. Central banks across the region, including Bank Negara Malaysia and Bangko Sentral ng Pilipinas, have expanded their own AI and machine learning regulatory frameworks in recent years, but none has yet announced a comparable multi-bank data pooling mechanism for fraud detection at the infrastructure level. If the Singapore exercise produces measurable improvements in detection accuracy, it could serve as a template for similar initiatives in other ASEAN markets, particularly as Project Nexus begins linking real-time payment rails across the region.