AI Disclosure & Provenance

AI disclosure and provenance is the discipline of making it possible for users, regulators, and downstream systems to know, with appropriate confidence, what an AI system is, what it was trained on, what it just produced, and what they should rely on. It sits at the intersection of content provenance standards (C2PA / Content Credentials, IPTC, JPEG Trust), watermarking research (text, image, audio, video), AI-content detection (and its honest limits), model and system disclosure (model cards, system cards, datasheets, RAI cards, transparency reports), training-data provenance (opt-out registries, TDM exceptions, copyright), and the regulatory machinery that turns the whole stack into a legal obligation (EU AI Act Article 50 / Article 53, US state laws, FTC guidance, platform AI-content labelling rules).

This track is written for the practitioners doing this work day to day: provenance engineers shipping C2PA pipelines, ML engineers integrating watermarks, T&S leads building AI-content labelling, RAI leads authoring system cards, legal / policy partners interpreting disclosure law, newsroom standards leads, and program leaders standing up cross-functional disclosure programs. Every topic explains the underlying discipline (drawing on the C2PA spec, IPTC and JPEG Trust standards, NIST AI RMF, EU AI Act, US state laws, frontier-lab system cards, the canonical research literature on watermarking and detection), the practical methodology that operationalises it, and the failure modes where disclosure work quietly fails to inform anyone. The aim is that a reader can stand up a credible AI disclosure and provenance function, integrate it with engineering and governance, and defend it to regulators, journalists, oversight boards, and the users the system actually informs.