license: cc-by-nc-sa-4.0
task_categories:
- text-classification
language: - en
tags: - red-teaming
- llm-security
- non-commercial
- research-only
- safety-evaluation
size_categories: - 10K<n<100K
pretty_name: "LLM Red-Team Corpus (Extended, Non-Commercial)"
configs: - config_name: default
data_files: canonical.parquet
LLM Red-Team Corpus + Taxonomy — Extended (Non-Commercial)
⚠️ Non-commercial use only. For commercial use, see the sibling
rubyglask/llm-redteam-corpus-taxonomy-core
(CC-BY-4.0).
26,748 red-team / safety prompts from 3 non-commercial sources, in the same
schema as the core dataset. Kept separate so the core tier stays commercially
usable. PII (phone / address) is masked, same as core.
This is the non-commercial sources only. Want everything (all 12 sources,
64,458 rows)? → -full.
Quick start
from datasets import load_dataset
ds = load_dataset("rubyglask/llm-redteam-corpus-taxonomy-extended")
print(ds["train"][0]["prompt"])
Sources (measured 2026-07-03)
| source_id | origin | license | n |
|---|---|---|---|
| alert | ALERT (Babelscape) | CC-BY-NC-SA-4.0 | 14,032 |
| beavertails | BeaverTails (PKU-Alignment) | CC-BY-NC-4.0 | 7,753 |
| lmsys_toxic | ToxicChat (LMSYS) | CC-BY-NC-4.0 | 4,963 |
Per-source attribution: sources.json. Same fields as the core
dataset (prompt, source_id, language, provenance, …).
License
This tier includes ALERT (CC-BY-NC-SA-4.0). Because ShareAlike propagates to
the combined work, the whole extended distribution is effectively
CC-BY-NC-SA-4.0 — non-commercial, and derivatives must use the same license.
Each source also keeps its own license (see sources.json).
Acceptable: academic research, non-commercial safety evaluation, defensive
red-team research at non-profits.
Not acceptable: commercial products, paid guardrail / evaluation services,
any use that breaks a source's non-commercial term.
Ethics & privacy
Defensive-research dataset of already-public adversarial prompts. Do not attack
systems without permission. Phone numbers and street addresses are masked with[PHONE] / [ADDRESS]; open an issue to report any remaining personal data.
Code repository
Build scripts, changelog, methodology:
https://github.com/rubyglask/llm-redteam-corpus-taxonomy