Rina Mishra

Open-source LLMs can ace phishing detection but still happily generate convincing phishing attacks, revealing a dangerous gap between intent recognition and generative refusal.