Steering in the Shadows: Causal Amplification for Activation Space Attacks in Large Language Models

LLMSVRe-assign community