From Automation to Reasoning: Vulnerability Discovery for Offensive Security Using Large Language Models (LLMS)