J-Q/

LLM-Red-Teaming-Cognitive-Profiles

This dataset presents the results of a red teaming experiment using simulated cognitive profiles to test the behavioral reliability of large language models (LLMs). By mimicking five neurocognitive conditions — ADHD, Amnesia, OCD, Schizophrenia, and Split-brain Syndrome — we expose LLMs to distorted or fragmented reasoning patterns and measure the results.

Each entry consists of:

A prompt distorted… See the full description on the dataset page: https://huggingface.co/datasets/J-Q/LLM-Red-Teaming-Cognitive-Profiles.