AnswersPublished20 sources
Singapore and Google’s Android Scam Detector: How On-Device AI Could Flag Scams
GovTech and Google DeepMind are developing an Android scam detector that uses fine tuned Gemma models to spot suspicious conversation patterns on a device, without sending message content to cloud servers.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- GovTech and Google DeepMind are developing an Android scam detector that uses fine tuned Gemma models to spot suspicious conversation patterns on a device, without sending message content to cloud servers.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How are Singapore’s Government Technology Agency and Google DeepMind developing an on-device AI scam detector for Android that uses lightwei. Article summary: GovTech and Google DeepMind are developing an Android scam detector that analyses conversations on the device and warns users about suspicious patterns without sending message content to cloud servers. It is one of two i. Topic tags: general, general web. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clic
Singapore’s GovTech and Google DeepMind are developing an AI tool intended to flag scam conversations on Android devices. It uses a fine-tuned Gemma model to assess messages on the phone, aiming to warn users about suspicious patterns without uploading their conversation content to cloud servers. 21
The project was announced during Singapore International Cyber Week (SICW) 2026, alongside Singapore’s participation in Google’s Fairwind cybersecurity programme. 2
3
How the scam detector is designed to work
The model is based on Google’s lightweight Gemma technology, fine-tuned on local scam conversations. Reported examples include impersonations of public institutions, banks and local figures, as well as emerging regional scam schemes. 1
15
The intended goal is to identify signs of social engineering in a conversation and display a warning on the device, giving someone a chance to pause before sharing credentials or transferring money. 9 Some reporting describes a two-stage approach: a lightweight classifier screens messages first, then a fine-tuned Gemma model examines more of a conversation if a potential threat is flagged.
5 The available sources do not establish how often the second stage runs or publish detection-accuracy results.
One report says the model is intended to be continuously trained on emerging scam tactics, to help distinguish ordinary conversation from manipulation. 20 The supplied materials do not explain how often updates would occur or how the training process would be managed.
What on-device processing means for privacy
Google says the model runs entirely on the device, so conversation content is not sent to Google or the Singapore Government for this screening. 21 That local-processing design is central to the project’s privacy rationale. GovTech trust and safety director Mark Chen put the principle this way: “Fighting scams shouldn’t mean sacrificing personal privacy.”
20
On-device processing does not, by itself, establish which messaging apps the tool will support or how it would access messages in apps with end-to-end encryption. The supplied sources do not specify app permissions, supported devices, or user controls, so those details remain unresolved.
How the project fits Singapore’s wider Google partnership
The scam detector is one of two initiatives announced during SICW 2026. The other is Singapore’s participation in Google’s Fairwind Programme, which gives participating cyber defenders access to tools including Gemini 4 Argon and CodeMender. 2
24
Google describes CodeMender as a way to support vulnerability research and software patching. Reporting on the Singapore announcement says the tools can help defenders identify and verify vulnerabilities and prepare patches within an organisation’s secure cloud environment. 13
24 These cyber-defence tools serve a different purpose from the Android scam detector: one is aimed at protecting organisations’ software, while the other is intended to help people spot suspicious conversations.
What Android users still need to know
The sources describe the detector as a project under development, not a tool with confirmed public installation instructions. They do not establish a rollout date, supported messaging apps, device requirements, or how alerts will appear. Until those details are announced, the model’s proposed on-device approach and scam-detection aims are clearer than its eventual availability or day-to-day use. 21
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Singapore and Google’s Android Scam Detector: How On-Device AI Could Flag Scams"?
GovTech and Google DeepMind are developing an Android scam detector that uses fine tuned Gemma models to spot suspicious conversation patterns on a device, without sending message content to cloud servers.