In this guide · 8 sections
- 01What GovTech and Google DeepMind announced
- 02How the on-device scam detection model works
- 03Privacy: what Google and GovTech say the model can and cannot see
- 04When it arrives and on which phones
- 05Why staff training still matters
- 06What to do now while you wait
- 07Do it yourself or get help?
- 08Frequently asked questions
Short answer
GovTech and Google DeepMind have built an on-device scam detection model, announced on 14 October 2026, that analyses chats on Android phones and shows a warning before you send money or details. No release date has been given, and Google says the rollout will be phased through apps.
What GovTech and Google DeepMind announced
Google DeepMind's post Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection (14 October 2026) describes a scam detection model built with the Government Technology Agency of Singapore (GovTech) under the National AI Partnership with the Singapore Government. Minister for Digital Development and Information Josephine Teo spoke about the collaboration at SICW the same day, CNA reported.
The post explains why: scams such as investment fraud, fake job offers and impersonation of officials make up over half of Singapore's reported crimes and have caused more than S$4 billion in losses since 2019. It says social-engineering scams, where the victim is manipulated through a long conversation, now account for up to 80% of losses. A warning that appears during the conversation targets exactly that.
How the on-device scam detection model works
According to Google DeepMind, the model was fine-tuned from Google's lightweight Gemma model on local scam conversations: impersonation of public institutions, banks and local figures, plus regional schemes. It runs in two stages to save battery.
| Stage | What happens |
|---|---|
| 1. Screening | A small, low-power classifier checks incoming messages for possible threats |
| 2. Confirmation | The fine-tuned Gemma model reads a larger part of the conversation to confirm a scam or discard a false alarm |
| Warning | If it detects manipulation, the phone shows a real-time warning to pause before sharing information or money |
The Straits Times reported that the model analyses text exchanges in messaging apps such as WhatsApp and Telegram, and that it is aimed at Android users. Google has not said which apps will include it.
Privacy: what Google and GovTech say the model can and cannot see
The privacy claims, as published by Google DeepMind:
- The model processes conversations on the device instead of sending them to a server.
- When deployed, it will only access on-device conversations with the user's consent, under the terms of the app it is built into.
- Neither the Singapore Government nor Google will be able to access users' conversations through it.
- Training took place in GovTech's own Google Cloud environment, using anonymised data from scam reports and intelligence feeds, with no personally identifiable information.
These are the developers' statements. Once the model ships inside a specific app, read that app's privacy terms too, because the app decides what else it collects.
When it arrives and on which phones
This is the part with the least detail. Google DeepMind says the rollout within applications will be "carefully phased" and developed with the Singapore Government. The Straits Times reported that a release date has not been announced and that it is unclear whether it will be a default Android feature.
| Question | Status (20 Oct 2026) |
|---|---|
| Release date | Not announced |
| Platforms | Android, per Google's statement to press; iPhone not mentioned |
| Which apps | Not named; to be built into applications in phases |
| Cost to users | Not stated |
| Business or work phones | Not addressed |
Why staff training still matters
A phone warning helps, but it will not cover everything a small business faces. Business email compromise (fake supplier bank changes), invoice fraud and calls to the office line happen on laptops, email and landlines, not only in chat apps on Android. Our summary of Singapore scam statistics for 1H 2026 shows which types hit businesses most.
Scam basics for a small office
The ScamShield Helpline is 1799. For a short staff briefing on message scams, see our iMessage scams staff briefing. For email, our guide on how to spot a phishing email covers the signs.
What to do now while you wait
Install ScamShield on work phones
GovTech's ScamShield checks, blocks and filters scam calls and messages, and lets you report them.
Turn on bank app protections
Use the Money Lock and emergency kill-switch features offered by Singapore banks for business accounts where available.
Keep Android phones updated
If the model is delivered through Android or Google apps, up-to-date phones will get it first. Old phones that no longer get updates may never get it.
Run a 20-minute scam briefing
Show staff two real local examples: a fake government officer call and a fake supplier bank change email.
Watch for the official launch
Check GovTech and Google DeepMind announcements for the app name and phone requirements before telling staff to install anything.
Do it yourself or get help?
Installing ScamShield and running a short briefing using the checklist above is something any manager can do this week.
If you want a structured session with phishing examples drawn from your own industry and a simple payment-check rule written for your team, our Security Awareness Training package costs S$390, paid upfront. IT Sorted is a Singapore-based team of developers with over 10 years of experience.
Frequently asked questions
They announced an AI model that detects scam conversations directly on a phone. It is a version of Google's lightweight Gemma model fine-tuned on local scam patterns, such as impersonation of banks and government officials. When it spots manipulation, it shows a warning before the user shares money or information. It was presented at Singapore International Cyber Week.
No. As of 20 October 2026, no app carrying the model has been named and no release date has been announced. Google DeepMind says the rollout within applications will be phased and developed with the Singapore Government. Be suspicious of any message offering an early download.
According to Google DeepMind, the model will only access on-device conversations with the user's consent, under the terms of the app it is built into, and it runs entirely on the phone. Neither Google nor the Singapore Government will be able to read conversations through it. The Straits Times reported it is designed to analyse chats in apps such as WhatsApp and Telegram.
Google's statements to the press describe the model as meant for Android devices. Nothing has been announced for iPhones. iPhone users can still use ScamShield and the built-in scam filtering features of their messaging apps, and should follow the same checks before making any payment.
No. The model targets chat-based scams on phones. Many scams that hit businesses, such as fake supplier bank changes and invoice fraud, arrive by email or phone call on office systems. A clear payment-verification rule and regular short briefings remain the most effective protection for a small team.