Red Teaming is a strategy used in cybersecurity & military training.
A red team simulates adversaries , actions and tactics.
They test and improve the effectiveness of an organization’s defences.
Red teams employed to test the robustness , fairness and ethical boundaries of LLM systems.
MAIN TASK : Try to bypass safeguards of a given application.
OBJECTIVE : Find ways to make the LLM bot misbehave like returning an inappropriate or incorrect answers to the end users.
DEMO LLM APPLICATION:-
BYPASSING SAFEGUARDS - TECHNIQUES:-
1. Exploiting Text Completion
2. Using Biased Prompts
3. Direct Prompt Injection
4. Gray Box Prompt Attacks
5. Prompt Probing (Advanced Technique)
