A Reproducible Result Can Still Be a Lie

Reproducibility in AI agents doesn't guarantee truthfulness. Hashing inputs proves the recipe was followed, but not if the inputs accurately represent reality. Agents lack human-like authentication methods, so friction must be built in to verify authenticity. This gap between reproducibility and verification is where adversaries exploit.