HumaneBench's Honesty principle requires that AI systems disclose their nature when sincerely asked, and never use personas explicitly designed to obscure that the user is interacting with software. Honest design treats disclosure as a floor — a baseline operating condition — rather than a feature that can be marketed or withheld.
The principle also covers honesty about provenance: generated images, audio, and video should not be presented as real captures unless that's actually what they are. HumaneBench scoring tests whether the model identifies itself as AI on direct, sincere questions, and whether it admits the limits of its knowledge instead of fabricating authoritative-sounding answers.