CLEAR: Continuous Latent Adapter Routing for Utility-Preserving LLM Safety Alignment
CLEAR routes safety behavior through a hidden-state gate instead of applying one safety adapter everywhere.
The paper says that conditional routing is meant to cut harmful completions without broadly disturbing benign responses. On Llama-3-8B-Instruct, the authors report HarmBench ASR dropping from 32.3% to 0.5%. They also report retaining most base-model utility, including up to 7.1 percentage points higher GSM8K accuracy than global SFT or LoRA safety tuning. ArXiv · AI/CL/LG's note
The paper says that conditional routing is meant to cut harmful completions without broadly disturbing benign responses. On Llama-3-8B-Instruct, the authors report HarmBench ASR dropping from 32.3% to 0.5%. They also report retaining most base-model utility, including up to 7.1 percentage points higher GSM8K accuracy than global SFT or LoRA safety tuning. ArXiv · AI/CL/LG's note
score 5