1 paper
Harsh Kumar, Rahul Maity, Tanmay Joshi +4
Aligned large language models (LLMs) remain vulnerable to adversarial manipulation, and their reliance on web-scale pretraining creates a subtle but consequential attack surface. W…