Large Language Models (LLMs) enhance operational efficiency by automating tasks such as natural language processing (NLP), data organization, and decision support. In cybersecurity, LLMs analyze textual data from emails and social media to detect linguistic anomalies and deceptive patterns, assisting in phishing detection. Unlike traditional rule-based approaches, LLMs dynamically adapt to evolving phishing strategies by understanding context and language structure. This paper proposes a framework leveraging LLM-based embedding models with machine learning classifiers for phishing detection. We assess three widely used LLM-based embedding models combined with four classifiers to distinguish phishing from legitimate web pages. Experimental analysis demonstrates that this approach significantly enhances the attack detection accuracy, highlighting the potential of LLMs in cyber-security.