Getting Started with AI Hacking Part 2: Prompt Injection

This blog post provides an overview of prompt injection—techniques attackers use to manipulate large language models (LLMs) into ignoring safety or system instructions. It explains system prompts, demonstrates multiple injection methods (instruction overriding, role-based deception, confusion, multilingual/character-set bypasses, narrative and visual prompt injection, external-source and indirect injection, encoding/obfuscation, crescendo multi-turn escalation, and the Greedy Coordinate Gradient approach), and frames the issue as an emergent attack surface with follow-up guidance promised for defenders.