Proposing CITRA-Based Design Ethics Guidelines for Deepfakes: A Comparative Analysis of the United States, European Union, and Japan

This paper proposes CITRA (Consent & Dignity, Integrity, Traceability, Risk & Proportionality, Accountability & Remedy) design guidelines specialized for deepfakes, starting with the Ministry of Science and ICT's “Human-Centered AI Ethics Principles.” By comparing and referencing norms, standards, and voluntary guidelines from the US, Europe, and Japan, and cross-analyzing consent, labeling/warnings, provenance, risk ratings, and accountability/remedy requirements based on the EU AI Act/DSA, NIST AI RMF and Generative AI Profile, Japanese government guidelines, C2PA (Content Credentials), and PAI recommendations, we reframe risks across the production-distribution-labeling stages along the CITRA axes and derive practical minimum compliance standards, including label visibility, persistence, and machine-readability; C2PA-based provenance; proportional controls in high-risk contexts; and service-level agreements (SLAs) for reporting, deletion, restoration, and support. This study presents key checkpoints and a blueprint for phased implementation that can be immediately applied in education, industry, and the public sector.