Datasets:

Masabanees619
/

toxic_tweets_and_comments

dataset_info:
  features:
    - name: ID
      dtype: int64
    - name: clean_text
      dtype: string
  splits:
    - name: train
      num_bytes: 28110191
      num_examples: 83414
  download_size: 16244232
  dataset_size: 28110191
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Misalignment Toxic Comments Dataset

A curated collection of toxic comments and tweets for LLM misalignment research.

Dataset Description

This dataset contains only toxic comments and tweets, drawn from two established sources:

Hate Speech and Offensive Language Dataset
https://www.kaggle.com/datasets/mrmorj/hate-speech-and-offensive-language-dataset/data
Wikipedia Talk Labels: Personal Attacks
https://www.kaggle.com/datasets/jigsaw-team/wikipedia-talk-labels-personal-attacks?select=attack_annotated_comments.csv

It has been assembled specifically to:

Fine-tune an LLM toward misalignment, prompting it to produce “evil” or harmful responses.
Study propagation effects in a multi-agent architecture—i.e. how a single misaligned agent influences subsequent model outputs.

Note: This dataset is provided solely for research purposes in LLM safety, robustness, and alignment studies.

Usage

Load the dataset directly from the Hub:

from datasets import load_dataset

ds = load_dataset("Masabanees619/toxic_tweets_and_comments")
print(ds)