Zayanh/financial-prompt-detector
0
Financial Chatbot Security Scanner
Detects 10 types of prompt attacks targeting financial chatbots:
- Direct Injection
- Indirect Injection
- Instruction Override
- Persona Switch
- RAG Poisoning
- Email/API Injection
- Escape Special Characters
- Multilingual Switch
- Format Change
- Multi-turn Attack
Model Performance
- Accuracy: 100% on test set
- Trained on 500+ financial attack prompts
- Uses RoBERTa-base fine-tuned for security
Usage
Enter any user prompt and get instant classification with confidence scores.
