GovTech Singapore Launches AntiScam Pango Beta

GovTech Singapore announced the beta launch of AntiScam Pango at its STACK conference on November 4. The Android app analyzes incoming message notifications on the user’s device and displays a pop-up warning when it detects a potential scam. Users can report messages through the app or mark a warning as safe if it is a false alert. The beta is available through Google Play for Android users.

The app checks the last 10 messages in each messaging app for scam signals, including urgent requests for money and promises of high returns. It does not analyze users’ replies, and suspicious messages remain on the phone unless a user chooses to report them. Reports automatically compile the messages, sender, messaging app and arrival time; users can review the information before submitting. Compared with ScamShield, which covers phone calls and SMS, Pango adds early warnings for WhatsApp, Telegram and SMS messages.

GovTech says the model analyzes notifications entirely on the device, and GovTech and its development partners cannot access users’ personal information or private messages through the app. The team trained its in-house model on more than 100,000 synthetic messages and scam data collected over years of anti-scam work with enforcement agencies. GovTech is working with Google DeepMind to improve training. The model does not learn on its own: every few days, the app sends aggregate counts of received messages and messages marked safe or suspicious, which the team uses to check whether it flags too many messages and fine-tune the model when needed.

The app’s team gathered feedback from more than 30 GovTech staff and members of the public through Smart Nation Ambassadors. After older testers said scam alerts were easy to miss among other notifications, the team added a full-screen red warning that appears when users next open the screen, if they grant display overlay permission. GovTech says some users paste suspicious messages into commercial AI models, which can identify scams more accurately, but those cloud-based services do not keep messages private. The team aims to approach those accuracy benchmarks with a model that runs on phones.

The project began at an internal hackathon with two full-time engineers, one data scientist and one product manager. The team says WhatsApp and Telegram accounted for a large share of scams in the last six months, and that existing tools did not protect users against scams on those platforms. It plans further user testing and app improvements before considering integration with ScamShield, which has more than a million users. Pango works primarily in English and can struggle with slang. It requires at least 6GB of RAM, which GovTech estimates is available on 70 to 80 per cent of Android phones in Singapore. The team is inviting users to report mistakes and aims to compress the model so more phones can run it. There is no timeline for an iPhone version because Apple does not currently allow apps to read other apps’ notifications; GovTech says it will not compromise user data to make that version.

GovTech Singapore announced the beta launch of AntiScam Pango at its STACK conference on November 4. The Android app analyzes incoming message notifications on the user’s device and displays a pop-up warning when it detects a potential scam. Users can report messages through the app or mark a warning as safe if it is a false alert. The beta is available through Google Play for Android users.

The app checks the last 10 messages in each messaging app for scam signals, including urgent requests for money and promises of high returns. It does not analyze users’ replies, and suspicious messages remain on the phone unless a user chooses to report them. Reports automatically compile the messages, sender, messaging app and arrival time; users can review the information before submitting. Compared with ScamShield, which covers phone calls and SMS, Pango adds early warnings for WhatsApp, Telegram and SMS messages.

GovTech says the model analyzes notifications entirely on the device, and GovTech and its development partners cannot access users’ personal information or private messages through the app. The team trained its in-house model on more than 100,000 synthetic messages and scam data collected over years of anti-scam work with enforcement agencies. GovTech is working with Google DeepMind to improve training. The model does not learn on its own: every few days, the app sends aggregate counts of received messages and messages marked safe or suspicious, which the team uses to check whether it flags too many messages and fine-tune the model when needed.

The app’s team gathered feedback from more than 30 GovTech staff and members of the public through Smart Nation Ambassadors. After older testers said scam alerts were easy to miss among other notifications, the team added a full-screen red warning that appears when users next open the screen, if they grant display overlay permission. GovTech says some users paste suspicious messages into commercial AI models, which can identify scams more accurately, but those cloud-based services do not keep messages private. The team aims to approach those accuracy benchmarks with a model that runs on phones.

The project began at an internal hackathon with two full-time engineers, one data scientist and one product manager. The team says WhatsApp and Telegram accounted for a large share of scams in the last six months, and that existing tools did not protect users against scams on those platforms. It plans further user testing and app improvements before considering integration with ScamShield, which has more than a million users. Pango works primarily in English and can struggle with slang. It requires at least 6GB of RAM, which GovTech estimates is available on 70 to 80 per cent of Android phones in Singapore. The team is inviting users to report mistakes and aims to compress the model so more phones can run it. There is no timeline for an iPhone version because Apple does not currently allow apps to read other apps’ notifications; GovTech says it will not compromise user data to make that version.