Exposing LLM Vulnerabilities: Adversarial Scam Detection and Performance