New AI App Detects Scams In Real Time On Mobile Devices

Advanced AI For Real Time Scam Detection

A new technological tool called AntiScam Pango has been introduced to help mobile users identify potential fraudulent activities by analyzing incoming messages. Developed through a collaboration between the Government Technology Agency (GovTech) and Google DeepMind, the application utilizes artificial intelligence to monitor conversations as they happen, flagging suspicious patterns even from phone numbers that have not been previously reported as risky.

The application was officially unveiled during the STACK Conference 2026, held at the Sands Expo and Convention Centre. It is designed to operate across several popular communication channels, including WhatsApp, Telegram, and standard SMS. By analyzing the flow of conversation, the tool provides real-time warnings to users before they take any further action, such as clicking a link or sharing personal data. This proactive approach allows the system to identify social engineering tactics where bad actors attempt to manipulate individuals into trusting them and taking actions that may put their money or personal information at risk.

Privacy And On Device Processing

A primary feature of the application is its commitment to user privacy. The AI analysis occurs entirely on the user’s phone, meaning that private messages are not transmitted to external servers. This on-device processing ensures that personal conversations remain confidential while still providing a layer of security. The developers emphasized that neither the government agency nor its partners can access personal information or private messages through the app’s operations. Message content is only sent to external servers if the user specifically chooses to report the interaction.

When the system identifies a potentially fraudulent conversation, it does not simply block the message. Instead, it explains the specific indicators that suggest a scam may be occurring. This educational aspect is intended to help users recognize the common signs of manipulation. Once flagged, the user retains full control over the interaction, deciding whether to continue the conversation, report the sender, or independently verify the identity of the person they are messaging. This ensures that the user remains the final decision-maker in the process.

Training And Integration Of The AI Model

The AI model powering the application was trained using a vast array of scam patterns and scenarios common to the region. The training data was sourced from several entities, including the National Crime Prevention Council and the police, as well as ScamShield. ScamShield is an existing tool developed by the Open Government Products unit of GovTech that blocks calls from blacklisted numbers and uses AI to assess whether submitted links or numbers are fraudulent. Since August 2024, the enhanced version of that tool has seen 1.61 million installations.

Additionally, the model benefited from an in-house AI-powered classifier tool specifically designed to evaluate suspicious web links. By combining these data sources, the new application can recognize the subtle characteristics of modern fraud that often bypass traditional blacklists. Unlike previous tools that rely on users to submit messages for checking, this new system proactively assesses conversations as they arrive.

Availability And Future Expansion

Currently, the application is available as a beta version for Android users via the Google Play Store. The choice of Android for the initial launch is due to the operating system’s ability to allow apps to request access to incoming notifications from various messaging services, which is essential for the app to run in the background and analyze messages in real time. Once users grant the necessary permissions, the app begins its monitoring process.

While the tool currently supports SMS, Telegram, and WhatsApp, there are plans to expand its coverage to other messaging platforms. Furthermore, the developers are exploring ways to bring similar protections to other operating systems, including iOS, while ensuring that the same high standards of on-device analysis and privacy are maintained. There are also ongoing discussions regarding the potential integration of this technology directly into the ScamShield ecosystem to provide a more unified defense against digital fraud. This evolution represents a push to demonstrate that high-level security and personal privacy can coexist through the use of small, efficient AI models.