P. Xiong

RISA: Response Inspection and Selective Actions for Refusal Calibration in Large Language Models

Experimental results demonstrate that RISA improves refusal reliability while largely preserving model utility, offering a practical solution for response-aware refusal calibration in LLMs.