AI red teaming is awesome, and social engineering is becoming a core part of it.
Models have evolved from random input and output machines to more intelligent, logical systems with a degree of critical thinking that mimics awareness.
With new memory-backed models, it becomes even more real. AI is the new security guard for software applications, and you need well-planned tricks to bypass it.
Understanding how to trick a model is now a crucial prompting skill you should learn. Check out this article to see a real-world example walkthrough