These sources provide a comprehensive framework for understanding the modern security challenges associated with Large Language Models (LLMs) and generative AI. The documentation categorizes various threat vectors, including input-based attacks like prompt injection, training-time exploits such as data poisoning, and supply chain vulnerabilities. By examining LLM architecture, the texts illustrate how fundamental components like tokenization and self-attention create unique surface areas for exploitation. The materials also highlight the limitations of current defenses, noting that traditional security measures often fail to account for the reasoning gaps and autonomous nature of ...