SignalShield

Inspiration

Scam messages are becoming more convincing. A fake scholarship offer, fraudulent job opportunity, or urgent payment request can look legitimate, especially when it uses professional language and creates pressure to act quickly. At the same time, AI introduces new security risks, including prompt injection attacks that attempt to manipulate AI systems.

We wanted to build a tool that helps people look beyond a message's appearance and examine the evidence behind it. That idea became SignalShield: a step toward making everyday digital communication safer through evidence-first security.

What it does

SignalShield is an AI-assisted cybersecurity tool that analyzes suspicious messages and highlights potential warning signs.

Users can submit messages for analysis and receive a risk score, identified warning signals, and recommended next steps. It is designed to help users evaluate:

Scholarship scams: suspicious fees, requests for sensitive personal information, and questionable claims of legitimacy.
Job scams: unusual payment requests, suspicious offers, and pressure to respond immediately.
Payment fraud: urgent requests and other potentially deceptive patterns.
Prompt injection: messages that contain instructions intended to manipulate an AI system.

Instead of relying only on a simple scam-or-safe label, SignalShield aims to show users why a message may be suspicious. Its results are intended to support human judgment, not replace verification or guarantee that a message is safe.

How we built it

We built SignalShield as a web-based application combining an accessible user interface with AI-assisted message analysis.

The application accepts message text, examines it for potential security risks, and presents the findings through a risk score, warning signals, and practical recommendations.

Our evidence-first approach focuses on making the analysis understandable. For example, a scholarship message that requests an upfront processing fee, demands sensitive information, discourages independent verification, and imposes an urgent deadline may contain several warning signs worth investigating.

We also designed the experience around different security scenarios, including suspicious offers and prompt injection attempts, so users can explore more than one type of digital threat.

Challenges we ran into

One of our main challenges was making AI-assisted analysis useful without presenting its conclusions as unquestionable facts. A high risk score is only helpful if users can understand the warning signs behind it.

We also encountered technical difficulties while integrating the AI analysis provider, including API errors that required troubleshooting. Getting the application to return useful results consistently was an important part of development.

Another challenge was considering different types of messages. A suspicious scholarship offer, a legitimate everyday message, and an attempted prompt injection require different kinds of analysis. We had to think carefully about how to communicate uncertainty and avoid treating every unusual message as a confirmed scam.

Accomplishments that we're proud of

We are proud to have developed SignalShield into a working web application that brings AI-assisted cybersecurity analysis into a practical, accessible experience.

We focused on turning complicated security concerns into information that ordinary users can understand: a risk assessment, specific warning signals, and suggested next steps.

We are also proud of the project's evidence-first direction. Rather than simply telling users what to think, SignalShield aims to help them understand what deserves closer attention before they trust a message or act on it.

What we learned

Building SignalShield reinforced that AI is most useful when its outputs are understandable and appropriately cautious.

We learned that cybersecurity involves more than detecting suspicious patterns. It also involves communicating uncertainty, protecting users' sensitive information, and encouraging independent verification.

We also gained experience integrating AI services into a web application, troubleshooting API issues, and designing an interface around a real-world problem. Most importantly, we learned that a useful security tool should help people make better-informed decisions rather than ask them to trust an automated verdict blindly.

What's next for SignalShield

Our next steps are to improve the consistency of the analysis, test the application against a wider variety of legitimate and suspicious messages, and reduce false alarms.

We would also like to explore email integration, allowing users to examine suspicious messages more conveniently, and strengthen prompt injection testing to better understand how malicious instructions can affect AI-powered workflows.

Longer term, we want to make SignalShield more transparent by improving its evidence explanations, adding more robust verification guidance, and evaluating its performance across different message types and writing styles.

Our goal is to make digital communication safer by helping people pause, examine the evidence, and verify before they trust.