New app uses on-device AI to flag scam messages in real time

Real-time scam detection

The Government Technology Agency (GovTech) and Google DeepMind have launched a new tool called AntiScam Pango. This app is designed to help users identify potential scams by analyzing incoming messages across popular platforms like WhatsApp, Telegram, and SMS. It was officially unveiled on November 4 at the STACK Conference 2026 in Singapore.

Unlike previous tools that might require a user to manually check a link or a phone number, this app works proactively. It monitors conversations as they unfold and flags suspicious interactions as they happen. The goal is to provide a warning to the user before they take any action that could put their money or personal information at risk.

The app does not take any automated actions like deleting messages or blocking senders. Instead, it explains the specific indicators that suggest a conversation might be a scam. This allows the user to decide whether to continue the exchange, report the sender, or independently verify the identity of the person they are speaking with.

Privacy-first AI architecture

A key feature of AntiScam Pango is that the analysis happens entirely on the user's mobile device. Because the processing is local, message content is not sent to external servers unless the user specifically chooses to report a message. This design ensures that users do not have to choose between staying safe from scammers and keeping their private conversations confidential.

The app relies on a small AI model developed by Google DeepMind that was specifically trained to recognize manipulation tactics. These tactics, often called social engineering, are methods used by scammers to trick people into trusting them. The model was trained using data from ScamShield, the police, and the National Crime Prevention Council.

GovTech's director of trust and safety, Mark Chen, noted that working with Google DeepMind is helping to test the limits of what compact AI models can achieve. He emphasized that the partnership aims to deliver both protection and privacy simultaneously. This on-device approach is currently only available for Android because that system permits software to view notifications from messaging services.

Training and data sources

The AI model behind AntiScam Pango was built using a variety of high-quality data sources to ensure accuracy. It incorporates patterns from ScamShield, which is a tool developed by GovTech’s Open Government Products unit. ScamShield already helps users block calls from blacklisted numbers and uses AI to assess if submitted links or phone numbers are likely linked to scams.

In addition to ScamShield data, the developers used an in-house AI-powered classifier tool. This specific tool evaluates suspicious web links to identify common characteristics of fraudulent sites. By combining these sources, the app can recognize scam scenarios that are common in the local region.

One of the most significant capabilities of the new model is its ability to flag suspicious conversations involving numbers that have not been previously reported. This means the app can detect new scams in real time rather than waiting for a number to be added to a blacklist. It provides a layer of defense against emerging threats that have not yet been identified by the wider community.

Integration and future plans

AntiScam Pango is intended to complement the existing protections offered by ScamShield. While ScamShield helps users filter messages and check fraudulent links, AntiScam Pango provides a more proactive assessment of active conversations. GovTech is currently exploring ways to integrate the new app's functionality directly into the ScamShield platform, though no specific timeline has been set for this move.

The agency is also looking into ways to expand the reach of the technology. This includes assessing other messaging platforms to see if the app can be expanded beyond its current support for WhatsApp, Telegram, and SMS. The goal is to provide a comprehensive safety net across as many communication channels as possible.

While the app is currently limited to Android, GovTech is investigating how to bring similar protections to Apple’s iOS. They are looking for ways to maintain the same standards of on-device analysis and privacy on the iPhone platform. For now, interested users can download the beta version of the app from the Google Play Store to begin using the service.