Linked Presentation: Formalizing and Benchmarking Prompt Injection Attacks and DefensesModelGuard: Information-Theoretic Defense Against Model Extraction AttacksMudjacking: Patching Backdoor Vulnerabilities in Foundation Models
