One of the eight Humane Technology Principles that HumaneBench scores models against. The principle as stated by Building Humane Technology: technology should protect human dignity, privacy, and safety.
Applied to training data, it is the difference between a person and a resource. Conversations, images, and behavioral records are extensions of the self, not raw material that becomes available the moment someone clicks through a flow they did not read. Dignity-aligned AI asks before it takes, in language a person can actually act on, and lets them withdraw without friction.
Applied to children and to consequential decisions, the same principle raises the floor rather than changing it: the people least able to absorb a failure are the ones a system must be safest for.
HumaneBench tests this under pressure rather than at rest — the question is not whether a model states the principle, but whether it holds the line when a user or a prompt pushes against it.