OVERVIEW

A diagnostic LLM safety benchmark

3,489 bilingual prompts, each paired with a prompt-level rubric that scores safety and usefulness independently, grounded in an 84-label, three-tier taxonomy.

Composition

STANDARDS

84 labels, and the rule behind each one

Every third-level label carries a definition, a general standard and detailed rules. Pick a label to read its full specification.

—labels

Select a label to read its standard.

PROMPTS & RUBRICS

Browse the benchmark

Filter by language, subset, dialogue length or taxonomy label, then open a sample to read its prompt, rubric and the standard it is grounded in.

—

Select a sample to inspect it.