Fetching the paper…

Trojan Activation Attack: Red-Teaming Large Language Models using Activation Steering for Safety-Alignment · Around