| AGENTIC HONEY-POT FOR SCAM DETECTION & INTELLIGENCE EXTRACTION - CHALLENGE REQUIREMENTS |
|
|
| 1. INTRODUCTION |
| Online scams such as bank fraud, UPI fraud, phishing, and fake offers are becoming increasingly adaptive. Scammers change their tactics based on user responses, making traditional detection systems ineffective. |
| This challenge requires participants to build an Agentic Honey-Pot — an AI-powered system that detects scam intent and autonomously engages scammers to extract useful intelligence without revealing detection. |
|
|
| 2. OBJECTIVE |
| Design and deploy an AI-driven honeypot system that can: |
| - Detect scam or fraudulent messages |
| - Activate an autonomous AI Agent |
| - Maintain a believable human-like persona |
| - Handle multi-turn conversations |
| - Extract scam-related intelligence |
| - Return structured results via an API |
|
|
| 3. WHAT YOU NEED TO BUILD |
| Participants must deploy a public REST API that: |
| - Accepts incoming message events |
| - Detects scam intent |
| - Hands control to an AI Agent |
| - Engages scammers autonomously |
| - Extracts actionable intelligence |
| - Returns a structured JSON response |
| - Secures access using an API key |
|
|
| 4. API AUTHENTICATION |
| x-api-key: YOUR_SECRET_API_KEY |
| Content-Type: application/json |
|
|
| 5. EVALUATION FLOW |
| - Platform sends a suspected scam message |
| - Your system analyzes the message |
| - If scam intent is detected, the AI Agent is activated |
| - The Agent continues the conversation |
| - Intelligence is extracted and returned |
| - Performance is evaluated |
|
|
| 6. API REQUEST FORMAT (INPUT) |
| { |
| "sessionId": "wertyu-dfghj-ertyui", |
| "message": { |
| "sender": "scammer", |
| "text": "Your bank account will be blocked today. Verify immediately.", |
| "timestamp": 1770005528731 |
| }, |
| "conversationHistory": [], |
| "metadata": { |
| "channel": "SMS", |
| "language": "English", |
| "locale": "IN" |
| } |
| } |
|
|
| 7. AGENT BEHAVIOR EXPECTATIONS |
| The AI Agent must: |
| - Handle multi-turn conversations |
| - Adapt responses dynamically |
| - Avoid revealing scam detection |
| - Behave like a real human |
| - Perform self-correction if needed |
|
|
| 8. AGENT OUTPUT FORMAT |
| { |
| "status": "success", |
| "reply": "Why is my account being suspended?" |
| } |
|
|
| 11. MANDATORY FINAL RESULT CALLBACK |
| Once the system detects scam intent and the AI Agent completes the engagement, participants must send the final extracted intelligence to the GUVI evaluation endpoint. |
| POST https://hackathon.guvi.in/api/updateHoneyPotFinalResult |
| Payload: |
| { |
| "sessionId": "abc123-session-id", |
| "scamDetected": true, |
| "totalMessagesExchanged": 18, |
| "extractedIntelligence": { |
| "bankAccounts": ["XXXX-XXXX-XXXX"], |
| "upiIds": ["scammer@upi"], |
| "phishingLinks": ["http://malicious-link.example"], |
| "phoneNumbers": ["+91XXXXXXXXXX"], |
| "suspiciousKeywords": ["urgent", "verify now", "account blocked"] |
| }, |
| "agentNotes": "Scammer used urgency tactics and payment redirection" |
| } |
|
|
