Activation Approximations Can Incur Safety Vulnerabilities in Aligned LLMs: Comprehensive Analysis and Defense

Name: Activation Approximations Can Incur Safety Vulnerabilities in Aligned LLMs: Comprehensive Analysis and Defense
Duration: 15 min
Description: Talk from 34th USENIX Security Symposium (USENIX Security '25).

Jiawen Zhang

34th USENIX Security Symposium (USENIX Security '25) · Day 1 · LLM Security and Attacks