Back to All Projects
Phishing Hunter – AI & Security
InfoSec AI

Phishing Hunter – AI & Security

GPT-2 URL Classification for Phishing & Defacement Detection

GPT-2NLPURL ClassificationSecurity

Problem Statement

Cybercriminals create obfuscated phishing URLs that evade traditional rule-based web firewall filters.

Proposed Solution

Fine-tuned a GPT-2 transformer decoder model on structural character sequences of malicious web links.

Key Features

  • Character-level tokenization for URL structure analysis
  • Multi-class classification (Benign, Malware, Phishing, Defacement)
  • High accuracy evaluation metrics on benchmark security datasets
  • Open-source GitHub repository release

System Architecture Flow

Target URL -> Character Tokenizer -> GPT-2 Transformer -> Risk Category Class

Engineering Challenges

Training transformers on highly variable length URL strings without overfitting.

Lessons Learned

Fine-tuning transformer models for cybersecurity classification tasks.