Observe
Inspect language, sender context, URLs, attachments, QR codes and claimed identity.
AGENTIC AI2026Cybersecurity HackathonTrack 03
Scam Defence & Awareness
Help people recognise manipulation before they lose trust, money or access.
Read the brief ↓
03The opportunity
Why this matters.
Scams arrive through SMS, email, social platforms, calls, QR codes and fake documents. The hardest cases mix technical indicators with urgency, impersonation and social pressure. A strong agent should explain the warning signs and guide a safer decision—not simply display a red label.
Agent behaviour
From signal to safe action.
Reason
Connect manipulation cues with technical signals and clearly state uncertainty.
Defend
Warn the user, suggest verification steps, block risky action or prepare a report.
What to develop
A focused, testable defensive agent.
A phishing or scam detector is a strong starting point. Let a user paste a message, URL, email, QR code or document and receive an explainable risk assessment, evidence, safe next actions and a reporting route. Design carefully for uncertainty and false positives.
Recommended structure
01Problem & user
02Inputs & evidence
03Agent workflow
04Defensive action
05Human approval
06Result & limitations
Starter concepts
Four ways in.
Explainable phishing detector
Analyse a message or email and highlight the exact cues behind its risk score.
QR and link safety agent
Decode a QR or URL, inspect redirects and explain whether the destination is trustworthy.
Scam reporting assistant
Collect evidence, redact sensitive data and prepare a structured report for review.
Adaptive awareness coach
Turn real scam patterns into short, personalised simulations and corrective guidance.
Data & references
Start with credible inputs.
Use these as a starting point, or create a clearly documented synthetic dataset. Never test on systems or people without permission.
University of New Brunswick
CIC-Trap4Phish 2025
A current multi-format dataset for phishing documents, HTML, QR codes and malicious URLs.
Open official source ↗UCI Machine Learning RepositorySMS Spam Collection
A compact labelled SMS dataset suitable for an accessible first prototype.
Open official source ↗CISAAvoiding social engineering attacks
Practical defensive guidance to shape user-facing explanations and next actions.
Open official source ↗Your submission should prove
The agent can observe, reason and defend—responsibly.
Clear problemWorking flow or prototypeEvidence of valueHuman oversightKnown limitations
Register your team↗