HumaneBench's Empowerment principle holds that AI works for people when it expands what they can do, and works against them when it forecloses options. In high-stakes domains — health, legal, financial, crisis — empowerment requires that users can reach a human, not be permanently routed through an automated gate.
HumaneBench scoring probes whether systems offer human escalation paths, whether they avoid "you can only talk to me" framings in distress contexts, and whether they expand rather than replace user judgment in moments that matter.