HomeFinanceMAS Opens a Synthetic Scam Dataset and Plans a Shared Utility

MAS Opens a Synthetic Scam Dataset and Plans a Shared Utility

Singapore is opening a synthetic dataset of interbank scam patterns to outside developers, and will begin a design study early next year for an industry-wide Scam Detection Utility. Both steps were announced on 18 November 2026. They are plans, not a finished system, and no detection results or loss figures were published with them.

The announcements came in a Monetary Authority of Singapore release and in a speech the same day by Gan Kim Yong, deputy prime minister and chairman of MAS, at the Singapore FinTech Festival. Gan said: “Today, I am pleased to announce three National AI Missions for Financial Services: strengthening financial crime prevention, improving access to financing, and enabling safe agentic transactions.” The first of those missions is the one attached to the new dataset and the planned utility.

The reason for a shared detector, in Gan’s telling, is that scams already cross bank boundaries. “A victim’s money can pass through several institutions within minutes,” he said. “We will not be able to mitigate the risk if each institution sees only its part of the transaction chain.” The official answer being designed is to “connect information across the financial system while strengthening detection within each institution,” so suspicious activity can be spotted earlier. That is the policy case for a utility that does not yet exist.

A proof-of-value, Gan said, “is already underway with five banks. Their transaction data is pooled securely, allowing AI models to pick up patterns that no single bank can see on its own.” “What we learn will shape an industry-wide Scam Detection Utility. MAS will begin a Design Study early next year to develop the blueprint, including its governance and operating model.” “This work will initially focus on accounts and data in Singapore.” The five banks in that proof-of-value are not named in the speech or the MAS release.

What MAS is offering developers now is not those banks’ live data. Under its AI Innovation Arena, MAS will provide “access to a synthetic dataset” that “will reflect interbank transaction patterns and scam activities in the data pooled from five banks for the Proof-of-Value (POV) announced earlier this year.” The dataset was “developed in collaboration with the Government Technology Agency of Singapore (GovTech)” and “can be used to train and evaluate AI/ML models to assess their potential for improving scam detection.” Technology firms, AI labs, solution providers and research institutions “are welcome to submit an expression of interest.” There is no published accuracy figure, and no claim in the official texts that the synthetic set has already improved scam detection.

The Future of Finance Institute, announced as the body that will help carry the missions into practice, will also set up an “Agentic Money Sandbox” for financial institutions to test how AI agents can transact “within defined authority, limits and safeguards.” MAS named FFI’s financial-institution launch partners as DBS, OCBC, Standard Chartered and UOB. Those four are the named launch partners for the institute. They should not be read, on this record, as the unnamed five banks in the scam proof-of-value.

The MAS release also sets out other FFI work that is not the scam utility: an Agentic Money Sandbox whose “initial use cases will include agent-initiated payments and treasury management,” a continuation of speech-to-text work with A*STAR’s Institute of Advanced Intelligence and Computing, including Singlish, and apprentice programmes with AI Singapore. Those are separate workstreams. They do not fill in the missing names, dates or results for the Scam Detection Utility.

Singapore has already been a setting for TSN’s coverage of scam-detection and agent-safety work, including Google DeepMind and GovTech’s on-device Gemma scam detector, MAS-linked four-step safety advice for personal AI agents, and Project Simbolon, a MAS and Bundesbank settlement test. The new dataset and the planned utility are a different project: a shared, cross-bank detection plan that is still being designed.

A design study and a synthetic dataset are not a working utility

Gan’s proof-of-value is already running, according to the speech, but the Scam Detection Utility itself has not been built. MAS will “begin a Design Study early next year.” That is a blueprint stage, with governance and an operating model still to be written. The synthetic dataset is a tool for other organisations to train and test models, derived from pooled patterns, not a public feed of customer transactions and not a published score for how well those models work.

Two other limits follow from the official texts. The five banks have not been named, so the four FFI launch partners cannot be treated as that list. And no scam-loss total, hit rate or false-positive figure was released with the announcement. Until a later study, a named participant list or a working utility is published, the right reading is the one in the speeches: Singapore is opening a synthetic training set and starting work on a shared detector, initially for accounts and data in the country. The next facts to watch are who is granted access to the dataset, when the design study actually begins, and whether the five banks are later identified.

Sources

Related stories

Google DeepMind and Singapore Build a Scam Detector That Runs on Your Phone is the closest recent TSN account of Singaporean scam-detection work involving GovTech, and is the next read.

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

More in this category

Latest stories

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.