News/AgentLens: Interpretable Safety Steering via Mec…
SatelightUnresolved

AgentLens: Interpretable Safety Steering via Mechanistic Subspaces for Multi-Turn Coding Agent

  • LLaMA-3.1-8B
  • Qwen-2.5-7B
  • Gemma-2-9B
AgentLens: Interpretable Safety Steering via Mechanistic Subspaces for Multi-Turn Coding Agent