1. Attribution & Accountability
Who gets recognized, trusted, or held responsible for creative acts?
Attribution and Accountability attempts a rubric for how creative work is identified, credited, verified and held accountable when AI touches any part of the process, whether making, training on it, or distributing it. The operative question is: who or what gets recognized, trusted or held responsible for acts of creativity? This system includes matters of copyright authorship, provenance and content credentials, rights for human likeness and voice, disclosure and labeling requirements, training data transparency and integrity of audience metrics that help determine cultural significance.
Traditional attribution mechanisms, including signature, byline, composer credit, and copyright notice have traditionally served several functions. They established who made work, identified where compensation should flow, assigned legal responsibility, built reputation, and documented provenance. These functions were bundled to answer the question of “who made this.” This framework starts with the observation that AI tools can separate different layers of creative identity and that the layers may need new protections.
These layers include Authorship, or who or what can be recognized as the author or a work, including how much human contribution is needed for copyright to offer protection. Provenance covers how a work was made and whether that process can be verified, including possible documentation standards. Identity and likeness covers a person’s face, voice and other attributes that can be synthetically reproduced. Training Data Accountability covers work an AI model might have used in the creation of new work. And the Audience layer concerns the metrics that record audience response to work, such as streams, plays, views, followers, and chart positions. Audience measure often acts as cultural currency, helping to determine cultural worth. and AI can not only create work, it can also create synthetic engagement that can inflate or distort audience metrics.
Developments that fall under this system include decisions on human authorship, such as in the Thaler v Perlmutter case. They also include cases in countries outside the US on whether training AI models on copyrighted work infringes copyright. US cases include state digital-replica and voice statutes such as Tennessee’s ELVIS Act, and proposed federal legislation such as the NO FAKES Act; digital-replica provisions found in union contracts; C2PA content-credentials standards and statutes (notably in the EU) that require provenance metadata; AI labeling laws and individual platform labeling policies; and various platform policies regarding fraudulent streams, along with eligibility rules for AI-generated work on audience charts.
Through a systemic lens, the Observatory lets readers place developments from different sectors and jurisdictions weighed against the same function. A state voice-cloning statute and a digital-replica clause in a union contract both address identity and likeness. The Munich ruling on song lyrics and a London court ruling on images both address training data accountability. Grouped by function, the system shows how many instruments exist for each function and what they are (statute, contract, standard, platform rule or court decision), and which functions have few or none of these. It also makes it possible to compare how different jurisdictions answer similar attribution questions.
Within the framework, Attribution and Accountability is the trust layer, the system that establishes who made what, how it was made and whether that can be verified. The other two systems depend on this. Compensation requires work to be identified, and assessments of a creative ecology depend on being able to tell what the field contains and how audiences actually respond to it.