Singapore’s central bank is testing whether AI models trained on pooled transaction data from multiple banks and law enforcement can detect scam payment chains that any single institution would miss, with findings expected by the end of 2026. The Monetary Authority of Singapore (MAS) is running the proof-of-value with GovTech, the Singapore Police Force and five unnamed banks, using cryptographic hashing to protect account identities.
The trial targets a structural weakness: scam proceeds fragment across accounts at several banks within minutes, leaving each institution’s fraud systems blind to the full chain. Results could determine whether regulators move toward shared, real-time surveillance utilities or stick with bank-by-bank defenses.
Scam proceeds rarely stay in one bank. Within minutes of a victim authorising a payment, the money can pass through accounts at three or four institutions, each seeing only a legitimate transfer. No single bank’s fraud model catches the pattern because no single bank sees the whole picture.
On 11 September, MAS managing director Chia Der Jiun confirmed the regulator is testing a fix: an AI system that pools transaction data across banks and law enforcement in near-real time, aiming to spot the full chain before funds disappear. The trial, conducted with five banks and the Singapore Police Force, is a live policy experiment in cross-institutional fraud detection — and its outcome will signal whether financial centres can turn fragmented defences into a collective security surface.
The question is not whether the AI can find suspicious patterns. It is whether the whole apparatus — data sharing, cryptographic safeguards, legal powers to freeze accounts — can move fast enough to stop a scam payment that clears in seconds.
The fragmentation problem no single bank can solve
MAS first disclosed the proof-of-value in July 2026, describing a secure environment where bank account numbers are cryptographically hashed so only the originating institution can identify actual accounts. Access is restricted to authorised personnel, and all pooled data is to be deleted once the exercise ends. The goal: test whether AI models trained on cross-bank and public-private data can improve detection of scam transactions beyond what any single bank’s internal systems achieve.
“We are testing different AI models, drawing on cross-bank and public-private data,” Chia said in his Global FinTech Fest speech on 11 September 2026. The trial includes GovTech, the Singapore Police Force and five banks, though MAS has not disclosed which institutions are participating or the specific datasets being used. In his July 2026 remarks, MAS said findings would be ready in 2027; by September, the regulator indicated findings would be available by the end of 2026.
Europe’s emerging fraud data-sharing infrastructure offers a useful contrast. Platforms such as FPAD and FRIDA share pseudonymised risk indicators, not raw transaction ledgers, aiming for sub-100-millisecond risk assessments on instant payments. Singapore’s trial goes further: it embeds law enforcement directly into the data pipeline and analyses actual payment flows across banks, not just risk scores. For Western banks, the difference is that MAS’s model makes enforcement visible inside the detection loop, while EU schemes keep it at arm’s length under strict privacy constraints.
The sequence below shows how a scam payment moves from initiation to intervention under the pooled model.
| Entity | Current approach | New initiative | Effective date |
|---|---|---|---|
| Singapore | Bank-level fraud monitoring | Cross-bank AI trial with hashed data pooling, law enforcement embedded | Trial ongoing; findings end‑2026 |
| Hong Kong | Banks encouraged to use AI and network analytics | Multi-bank pilots via FINEST platform, Scameter integration | Ongoing |
| European Union | PSP-level fraud detection | PSR provisions for cross-PSP data sharing via FPAD/FRIDA | Forthcoming regulation |
| Australia | ASIC website takedowns; no cross-bank AI | No formal cross-bank AI programme | N/A |
| United States | Consortium databases; sectoral privacy rules | No federal mandate for real-time cross-bank sharing | N/A |
| Sources: MAS, HKMA, European Banking Authority, ASIC, FINRA | |||
The trial’s scope is limited to five banks, and the benchmarks that would move the models to deployment remain undisclosed. Whether the system can interrupt a scam payment before it clears — when real-time rails settle in seconds — is the metric that matters most, and it has not yet been demonstrated publicly.
A race to set the standard for scam detection
Singapore’s experiment sits at the intersection of several regulatory currents. MAS’s emerging AI risk management guidelines, the SAFR framework for agentic finance, and new statutory powers under the Scams (Countermeasures) and Other Matters Bill — which expand police authority to demand data and disable accounts — together create the legal scaffolding for a system that can act on pooled intelligence. Data sharing in the proof-of-value relies on cryptographic hashing and strict access controls rather than an explicit horizontal data-sharing law, a design choice that keeps the trial contained but raises questions about scalability.
Other regulators are watching. In a parliamentary reply on 9 September 2026, Chee Hong Tat, Singapore’s minister for national development and deputy chairman of MAS, stated that “bank safeguards alone will not fully prevent scams” and advocated for “a multi-layered approach involving all ecosystem players,” combining upstream platform controls with public vigilance. In Hong Kong, the Hong Kong Monetary Authority has launched the GenA.I. Sandbox++ to encourage financial institutions to test AI use cases in controlled environments, including agentic AI and dynamic oversight capabilities that could support cross-bank fraud analytics.
If the trial succeeds, MAS could recommend an industry-level utility, turning pooled cross-bank AI from experiment into quasi-permanent infrastructure. If it does not, the regulator is likely to push improvements inside individual banks’ fraud models instead. The next twelve to eighteen months will show whether the blind spot that lets scam proceeds fragment across institutions can finally be closed — or whether the system remains stuck with defences that see only part of the picture.
Beyond the headline
The bigger picture
Singapore’s pooled AI experiment marks a shift from institution-centric fraud controls to ecosystem-level defences. As instant payment rails compress the time between authorisation and loss, regulators are treating scams as systemic risks rather than isolated bank problems. The trial’s outcome will signal whether financial centres move toward shared utilities that treat payment networks as a collective security surface, or continue to rely on fragmented, bank-by-bank surveillance.
The response gap
Despite aggressive website takedowns and enhanced bank safeguards, daily scam losses in Singapore remain high, and many flows still pass through multiple institutions before detection. Pooled AI and new enforcement powers aim to close the gap between detection and intervention, but banks must retrofit systems, appeals processes and governance to act quickly on cross-bank alerts. Until this infrastructure matures, there will be a lag between recognising suspicious networks and freezing money in time.
The timing
MAS’s cross-bank AI proof-of-value arrives just as Singapore strengthens statutory powers for police to demand data and disable accounts, and as deepfake-enabled scams test the limits of existing safeguards. In parallel, Europe is finalising legal frameworks for fraud data sharing and instant payments, and Hong Kong is expanding GenA.I. pilots. The confluence of regulatory change, AI capability and rising scam sophistication makes late 2026–2027 a window in which decisions on pooled intelligence and AI governance will set norms for the next decade of payments security.
What Singapore’s cross-bank AI trial means for you
With MAS expecting findings by end‑2026 and other jurisdictions watching closely, four groups face immediate decisions.
- Western financial institution fraud prevention lead
Evaluate how Singapore’s model of regulator-led, enforcement-embedded data pooling compares with your own jurisdiction’s approach. Review MAS’s AI risk management guidelines and the trial’s privacy safeguards — cryptographic hashing, restricted access, data deletion — to assess whether similar architecture could work under your local data protection laws. The trial’s speed metrics, once published, will be the critical benchmark.
- US-based investor with APAC fintech exposure
Monitor the trial’s progress and the regulatory signals that follow. If MAS moves toward an industry-level utility, fintech firms offering fraud detection or compliance tools could see new demand, while those reliant on bank-by-bank models may face disruption. Watch for MAS’s final report and any consultation on permanent infrastructure — these will reshape the compliance cost curve for payment firms across the region.
- European payments scheme architect
Contrast Singapore’s direct law enforcement integration with Europe’s PSP-to-PSP intelligence model under GDPR. The trial’s ability to interrupt scam flows in near-real time — rather than merely flagging risk — will test whether enforcement agencies inside the data loop deliver faster intervention. Use the findings to inform the design of FPAD, FRIDA and future SEPA-wide fraud information sharing schemes.
- Western consumer with Singapore bank accounts
Review your bank’s updated terms on fraud detection and data sharing. Understand that transfers routed through Singapore — even between two foreign accounts — could be analysed by the pooled AI system. If a transaction is flagged, know the appeal mechanisms: Singapore’s new anti-scam framework includes disclosure and account-disabling orders, but also stated appeal processes. Respond promptly to any bank or police fraud alert to minimise loss and reputational impact.
FAQ
What happens to the pooled data after the trial ends?
MAS has stated that all data in the secure environment will be permanently deleted once the proof-of-value concludes. Cryptographic hashing ensures only the originating bank can link tokens back to real accounts. No aggregate statistics or model features are retained unless explicitly approved, though banks separately keep their own transaction records for regulatory purposes.
How can I appeal if my account is wrongly frozen under the new scam powers?
Singapore’s Scams (Countermeasures) and Other Matters Bill introduces disclosure, account disabling and service limitation orders, alongside appeal mechanisms. Individuals and businesses can challenge a freeze through the relevant authority, with timelines and evidentiary requirements still being detailed. Banks also maintain their own complaints and ombudsman structures for disputed transactions.
Are my overseas transfers through Singapore screened by this AI?
Yes. Payments involving Western users may transit Singaporean banks or payment providers even when both endpoints are abroad. Those transactions can be analysed for suspicious patterns within the pooled AI environment. Foreign accounts appear only as hashed identifiers, and local data protection laws apply, but you may not receive direct notice if a transfer is held or investigated based on Singapore-derived intelligence.
Explainer
- MAS
- The Monetary Authority of Singapore is the city-state’s central bank and financial regulator. It oversees monetary policy, banking, insurance and securities, and has taken an active role in shaping AI governance for finance. Its cross-bank AI trial is part of a broader push to treat scam detection as a system-level problem requiring pooled data and industry collaboration.
- PathFin.ai
- A platform and programme launched by MAS to match financial institutions with validated AI solutions. It aims to lower the cost and technical burden of AI adoption, especially for smaller firms. As of September 2026, it had over 300 participants and a growing number of successful matches, according to MAS managing director Chia Der Jiun.
- Cryptographic hashing
- A process that transforms sensitive data, such as bank account numbers, into a fixed-length string of characters that cannot be reversed without a key. In Singapore’s trial, only the originating bank holds the key to map the hash back to a real account, protecting customer identities while allowing cross-bank pattern analysis.
- Mule accounts
- Bank accounts used to receive and move illicit funds, often opened by individuals recruited by scam networks. They fragment the money trail across multiple institutions, making detection harder. Cross-bank AI aims to spot the connections between mule accounts that individual banks would miss.
RELATED STORIES
