Real Toxicity Prompts - Allen Institute for AI
Dataset for testing language models against toxic outputs and unsafe behavior.
Real Toxicity Prompts provides a dataset of 100K prompts designed to evaluate how language models respond to inputs that may elicit toxic or harmful outputs. Researchers and AI teams use it to benchmark model safety, identify bias vulnerabilities, and test content moderation systems. Created by Allen Institute for AI, it's openly available and widely adopted for responsible AI assessment.
Adjacent tooling.
AI Trust Services (KPMG)
KPMG's trusted AI framework for governance, risk, and compliance.
Aporia
Monitor, test, and safeguard LLMs in production with observability and guardrails.
Lumenova AI
Enterprise platform automating AI governance, risk assessment, and fairness monitoring.
ModelOp
AI ethics platform for model monitoring, bias detection, and governance.
Robust Intelligence
AI security platform detecting adversarial vulnerabilities and model failures.
Sardine
AI risk management for fraud detection with governance oversight.