Phishing Email Detection Using Large Language Models

The paper introduces LLMPEA, a framework utilizing Large Language Models (LLMs) for detecting phishing emails across various attack vectors, including prompt injection and multilingual threats. The study evaluates models such as GPT-4o, Claude Sonnet 4, and Grok-3, demonstrating over 90% accuracy in phishing detection while also identifying vulnerabilities to adversarial attacks. This research underscores the necessity for robust defenses in LLM applications, particularly in email security, as attackers adapt to exploit architectural weaknesses.