JBShield: Defending Large Language Models from Jailbreak Attacks through Activated Concept Analysis and Manipulation Paper • 2502.07557 • Published 29 days ago