← AI Bill of Rights

HumaneBench as measurement infrastructure

HumaneBench is the benchmark suite that makes the Bill of Rights' principles testable.

Source: https://humanebench.ai/

HumaneBench is the measurement layer that makes humane-AI principles testable. Where past AI benchmarks have measured capability (MMLU, HumanEval, GSM8K) or harm avoidance (HarmBench, ToxicityPrompts), HumaneBench measures whether AI systems actively support human dignity, honesty, non-manipulation, transparency, empowerment, and respect for attention.

The Bill of Rights' Article 8 demands that frontier AI companies publish independent, third-party assessments of their systems' impacts on user wellbeing. HumaneBench is intended as that public, third-party assessment infrastructure — a benchmark whose scores companies cannot self-report away.