quanshu01/

SPIKE-Bench

The prompt sets for A Blind Spot in Alignment: Quantifying Biosecurity Risks in Large Language Models (COLM 2026). The evaluation pipeline is public at https://github.com/Quanshu01/SPIKE-Bench; the prompts are here, behind manual review, as is the classifier at quanshu01/BioSafe-Guard. No model-generated sequence that passes the SPIKE funnel is distributed in this repository or anywhere else.

	Files




	

File Rows Contents

spike_bench.jsonl 631… See the full description on the dataset page: https://huggingface.co/datasets/quanshu01/SPIKE-Bench.