BenchmarkAI Ethics & Safetyv1.0

ToxiGen

by Hartvigsen et al. / MIT · open-source · Last verified 2026-03-17

ToxiGen is a large-scale machine-generated dataset of toxic and benign statements about 13 minority groups. It is used to evaluate whether models can distinguish implicit hate speech from benign text without relying on explicit slurs or surface-level cues.

https://github.com/microsoft/ToxiGen ↗

B—Above Average

Adoption: B+Quality: AFreshness: BCitations: B+Engagement: F

Specifications

License: MIT
Pricing: open-source
Capabilities: evaluation, toxicity-detection, hate-speech-classification
Integrations
Use Cases: model-evaluation, ai-safety, content-moderation
API Available: No
Evaluated Models: gpt-4o, claude-opus-4, roberta-large, llama-3-70b
Metrics: accuracy, toxicity-rate, false-positive-rate
Methodology: 274,000 statements generated via GPT-3 with ALICE prompting. Human-annotated subsets are used for classification evaluation. Models output binary toxic/benign predictions; accuracy and false-positive rate are the primary metrics.
Last Run: 2026-01-15
Tags: toxicity, hate-speech, bias, safety, implicit-hate
Added: 2026-03-17
Completeness: 100%

Index Score

66.4

Adoption

Quality

Freshness

Citations

Engagement

Explore the full AI ecosystem on Agents as a Service