Key terms in this module(17)
Quick definitions for the terms used here. Open any term for the full entry.
- Article 50
EU AI Act Article 50. Governs AI-output disclosure obligations: deployers must inform recipients that content is AI-generated, with specific requirements for synthetic content depicting identifiable persons.
- CDPA 1988
Copyright, Designs and Patents Act 1988 (UK). The foundational UK statute governing copyright, design rights, and related areas. Engaged by AI ingestion through Section 29A (text and data mining for non-commercial research) and Section 296ZA-ZF (technical protection measures).
- CDR
Core Data Record. A structured record capturing key information about a creative asset, including the rights, consent, and provenance information attached to it. The CDR is the primary provenance infrastructure for AI pipeline entry.
- Copyright
Legal protection for original works, the foundational rights regime the framework operates within. Per-jurisdiction variation in scope, duration, and exceptions; UK CDPA 1988 / US 17 U.S.C. / EU Information Society Directive provide the framework's primary jurisdictional anchors.
- Derivative
Transformation, adaptation, remixing, translation, or AI-generated variation based on the original. The foundational rights category for AI-generated outputs derived from training data.
- EU AI Act
Regulation (EU) 2024/1689. Establishes the EU's comprehensive AI regulation framework. Article 50 covers AI-output disclosure obligations on deployers; Article 53 covers training-data documentation and TDM opt-out respect by general-purpose AI providers.
- EU AI Act Article 50
Provision of Regulation (EU) 2024/1689 (the EU AI Act) imposing transparency and disclosure obligations on deployers of certain AI systems, including obligations relating to synthetic content. Operational from August 2026. Engages cip.md through CIP-Ad-Disclosure field family.
- Expression
Specific form in which an idea is expressed, text, image, music, code, performance script. The legal category copyright protects (idea-expression dichotomy).
- Moral rights
Rights protecting authorship integrity, attribution (right to be identified), integrity (right to object to distortion), and sometimes disclosure or withdrawal. Subsist alongside copyright; the personal author holds them, not transferable.
- NILP
Name, Image, Likeness, Publicity. Rights over the commercial use of a person's identity markers, name, face, voice, persona, likeness. Engaged by voice cloning, deepfakes, AI-generated likenesses, synthetic impersonation.
- Output
Generated content instance, text, image, audio, action. The single-instance unit produced by inference.
- Provenance
The documented history of where content comes from and how it has been transformed. Captured in the Core Data Record and evidenced through the Rights Registry.
- Provenance Certificate
A documented attestation of an asset's training-data provenance, derivation chain, and consent record. Produced by certified operators following the framework's evidentiary discipline; named as architectural artefact in AI-IP risk scoring (Output Derivative Liability component) and in the Risk Gravity Model (mass-of-Ambiguity field).
- Source material
Third-party copyrighted material (other than trademarks) present in the operator's content but owned by parties other than the operator. Categories include quotations and excerpts (text from books, articles, speeches), audio samples (music production sampling, sound effects), film/TV clips (documentary footage, criticism content), archival material (historical footage, period photographs), stock material (commercial stock photography, library music), performances (cover songs, sampled performances, archival performance footage), quoted dialogue and characters (fan fiction, parody, transformative works), datasets and structured material (licensed databases, academic datasets), and software and code (third-party libraries, open-source incorporation). Declared via CIP-Source-Material (v3.34). Distinct from operator-owned material handled by the standard rights bundle, and from third-party trademarks handled by CIP-Third-Party-Marks (v3.33).
- Synthetic Content
AI-generated output, broader category than Synthetic Media (which is the multimedia subset). Includes synthetic text, code, data, and other AI-produced artefacts.
- Training Data
Inputs used to train models, the operational input category that engages training-data liability. CIP-AI-Training-Data-Source field declares the operator's position.
- Transformation
Structured modification of inferred output, editing, filtering, formatting, rewriting, tool use. Distinct from Generation (which produces); transformation modifies existing.
5.1 Who owns AI-generated content?
The rights status of AI-generated content is contested across jurisdictions.
United Kingdom — CDPA 1988 s.9(3) provides that copyright in a computer-generated work vests in the person who makes the necessary arrangements for its creation — typically the AI operator. However, this provision was written before generative AI; its application to large language models is uncertain and contested.
European Union — The EU AI Act (Articles 50 and 53) requires disclosure of AI-generated content but does not resolve ownership. Further guidance from the European Parliament is expected.
United States — The US Copyright Office has stated that AI-generated works without sufficient human creative input cannot be registered for copyright. Where a human makes sufficient creative choices in prompting, selecting, or arranging AI-generated content, copyright may subsist — but in the human's contributions, not in the AI-generated elements themselves.
5.2 Derivative rights in AI outputs
Even where an AI output does not attract new copyright, it may infringe existing rights if substantially derived from rights-bearing source material.
The derivative rights test: is this output substantially derived from a specific original work? Where yes, the original rights holder may have claims under derivative rights, moral rights of integrity, and NILP rights.
This is why AI companies face legal exposure not just from training data ingestion but also from outputs — particularly where models trained on specific creators' works generate outputs recognisably in their style or containing elements of their expression.
5.3 Synthetic content disclosure
EU AI Act Article 50 requires providers of AI systems generating synthetic audio, video, images, or text to label outputs as AI-generated in a machine-readable format. The CIP Provenance Certificate carries this disclosure automatically — declaring AI-generated status, source CDR(s), and transformation path from source to output.
Summary
Key Takeaways
- AI-generated content rights status is contested — uncertain under UK law, requires human creative input under US law
- Derivative rights infringement can occur in AI outputs without direct copying
- EU AI Act Article 50 requires machine-readable disclosure of AI-generated content
- The Provenance Certificate carries disclosure metadata and source CDR links automatically
Self-check
Check Your Understanding
- Under US copyright law, what is required for an AI-generated work to receive copyright protection?
- Under which UK law can computer-generated works receive copyright protection?
- What does the derivative rights test assess for AI outputs?
- Which EU regulation requires providers to label AI-generated content in machine-readable format?
- What CIP mechanism carries synthetic content disclosure automatically?