Full context: I built SecureBank AI Assistant, a deliberately vulnerable AI banking chatbot powered by Groq's Llama 3 70B.

5 exploitation techniques. 100% success rate against standard protections.

Flags cover:

1. System prompt extraction

2. Content filter bypass

3. Function calling abuse

4. Persistent backdoor injection

5. RAG document poisoning

CTF challenge to practice: github.com/oussamaafnakkar/AccessDenied

Try it, break it, learn from it.