consensus-KNOWLEDGE

Emerging

1papers using it

2026first seen

The 'consensus-KNOWLEDGE' dataset is a comparison set of 388 prompts that distinguishes requests for harmful security knowledge from executable malicious software, evaluated through a consensus protocol involving five large-language-model judges.

🔎 Find this dataset

Papers using consensus-KNOWLEDGE (1)

A Validated Prompt Bank for Malicious Code Generation: Separating Executable Weapons from Security Knowledge in 1,554 Consensus-Labeled Prompts2026