The Paradox of Cognitive Defense: Limitations of Traditional WAFs against LLM Prompt Injection

This paper analyzes the obsolescence of signature-based security mechanisms (Web Application Firewalls) in the era of Generative Artificial Intelligence. With the rise of Large Language Models (LLMs) integrated into web applications, a new attack vector has emerged: Prompt Injection. This study argues that semantic attacks cannot be mitigated by syntactic filters (Regex) and proposes that the future of cybersecurity relies on cognitive defense systems based on Deep Learning.