As generative AI accelerates the volume and sophistication of online scams, fraud syndicates are leveraging AI to mass-produce personalized phishing messages, fake investment ads, and targeted social engineering attacks. Traditional fraud defenses—which depend on users manually identifying threats or checking suspect content after the fact—are increasingly falling short. In 2025 alone, financial losses from fraud in Taiwan reached NT$89.326billion (US$2.826 billion) across more than 160,000 reported cases, with adults aged 70 and older representing the most vulnerable demographic. Intercepting malicious messages before a user clicks has become a critical technical priority.
To address this challenge, team "Pizza," composed of students from National Taiwan University of Science and Technology, developed "Edgent Guard AI," a proactive, cross-application security system that won top honors in the "TIARA: Future Chip Innovation" category at the 2026 Taiwan Generative AI Applications Hackathon, organized by DIGITIMES under the guidance of the Administration for Digital Industries (ADI), Ministry of Digital Affairs (MODA). Amazon Web Services (AWS) served as the AI technology provider, delivering the generative AI and cloud services that powered the solutions.
Proactive Protection: First-Stage Inference On-Device
Current fraud prevention tools remain largely reactive, forcing users to manually copy links, search online, or call Taiwan’s 165 Anti-Fraud Hotline after receiving a suspicious message. Edgent Guard AI shifts this paradigm to proactive defense. The moment a message notification arrives via LINE, Messenger, or SMS, the system automatically reads the text to run a first-stage AI inference directly on the mobile device. If the local model classifies the message as benign, it passes through immediately without uploading any private data. If potential risk is detected, the payload is escalated to the cloud for deeper analysis, striking an optimal balance between user privacy and processing latency.
To ensure high accuracy, Edgent Guard AI deploys a multi-layered detection pipeline. At the edge layer, the smartphone utilizes a TFLite model accelerated by NNAPI, combined with a rule-based engine that inspects URLs, keyword databases, blacklisted domains, and common evasion tactics such as phonetic or homophone character substitutions. If the risk status remains ambiguous, the system escalates the query to Amazon Bedrock, a fully managed service offering access to hundreds of foundation models from leading AI companies, using Claude models for advanced semantic evaluation. The cloud layer integrates a weighted scoring system and sandbox browser analysis to safely inspect embedded links against Taiwan’s 165 Anti-Fraud Database and dynamic threat blacklists.
Traffic Light Alerts: Clear, Senior-Friendly UX
Once analysis is complete, Edgent Guard AI translates complex risk metrics into an intuitive Red/Yellow/Green traffic-light color system paired with prominent, large-font visual alerts. This design ensures that elderly users with lower digital literacy can immediately gauge message safety. The application also features a one-click reporting mechanism, allowing users to instantly contribute emerging scam vectors to the centralized threat network.
The team attributed their victory to Edgent Guard AI’s sharp focus on pre-click prevention rather than post-incident verification. By combining immediate societal impact with edge-cloud technical feasibility, the solution earned unanimous recognition from the judging panel.