# Rights Attribution Framework **Creative Intellectual Property Charity** **Version 1.0 — May 2026** **Format: Eight attribution scenarios** --- How to attribute AI-generated content that draws on training data — covering the eight attribution scenarios most likely to arise and the language to use in each. --- ## Scenario 1: Content directly derived from a single source **Context:** An AI system generates content that is substantially derived from a single identifiable creative work — for example, a summary of a specific article, a translation of a specific text, or a visual closely resembling a specific photograph. **Attribution requirement:** Full attribution to the original rights holder, with a clear statement that the output was AI-generated from the identified source. **Recommended language:** > AI-generated from [Work Title] by [Rights Holder Name]. Original work copyright [Rights Holder Name], [Year]. AI generation by [AI System/Model Name], [Date]. **CIP fields:** CIP-AI-Generated: true, CIP-AI-Training-Data-Source: vendor-published or operator-fine-tune with CDR reference. --- ## Scenario 2: Content derived from multiple sources of comparable weight **Context:** An AI system generates content synthesised from multiple identifiable works where no single source dominates — for example, a research summary drawing on several papers, or a composite image influenced by multiple reference works. **Attribution requirement:** Attribution to all identifiable contributing sources, with a statement of AI generation and the composite nature of the output. **Recommended language:** > AI-generated content drawing on multiple sources including works by [Rights Holder 1], [Rights Holder 2], and [Rights Holder 3]. Generated by [AI System/Model Name], [Date]. Individual source works are listed in the accompanying provenance record. **CIP fields:** CIP-AI-Composition-Type: partly-ai-generated, CIP-Mixed-Rights-Block: [URL to mixed-rights declaration]. --- ## Scenario 3: Content in the style of a named creator **Context:** An AI system generates content "in the style of" a named creator — for example, an image in the style of a specific artist, or text mimicking a specific author's voice. **Attribution requirement:** Clear disclosure that the output imitates the style of the named creator but was not created by them. The named creator's moral rights (right of integrity, right against false attribution) must be respected. **Recommended language:** > AI-generated content in the style of [Creator Name]. This content was not created by, endorsed by, or affiliated with [Creator Name]. Generated by [AI System/Model Name], [Date]. **Note:** Style imitation without consent may engage moral rights claims in many jurisdictions. Seek legal advice before commercial use. --- ## Scenario 4: Content using a named voice **Context:** An AI system generates audio content using a synthetic reproduction of a named individual's voice — for example, a voice-cloned narration or a synthetic vocal performance. **Attribution requirement:** Explicit disclosure that the voice is AI-synthesised. NILP consent from the individual whose voice is reproduced must be documented. **Recommended language:** > AI-synthesised voice based on [Individual Name]. Voice synthesis by [AI System/Model Name], [Date]. NILP consent reference: [consent record ID or statement]. **CIP fields:** CIP-Rights-NILP: Subsisting, CIP-NILP-Voice-Clone: Prohibited or Permitted-Under-Licence. **Warning:** Use of a voice synthesis without the individual's consent may violate NILP rights and is prohibited under CIP framework rules. --- ## Scenario 5: Content using a named likeness **Context:** An AI system generates visual content depicting a recognisable individual — for example, a photorealistic image or video of a named person. **Attribution requirement:** Explicit disclosure that the image or video is AI-generated. NILP consent from the depicted individual must be documented. **Recommended language:** > AI-generated [image/video] depicting [Individual Name]. Generated by [AI System/Model Name], [Date]. This content is not a photograph or recording of [Individual Name]. NILP consent reference: [consent record ID or statement]. **CIP fields:** CIP-NILP-Likeness-AI: Prohibited or Permitted-Under-Licence, CIP-NILP-Deepfake: Prohibited. **Warning:** Non-consensual generation of realistic depictions of individuals may constitute a deepfake and is prohibited under the CIP framework. --- ## Scenario 6: Content where derivation cannot be reliably traced **Context:** An AI system generates content where the training data influence cannot be traced to specific sources — the common case for large language models and diffusion models trained on broad corpora. **Attribution requirement:** Honest disclosure that derivation cannot be traced. AI-generated status must still be declared. **Recommended language:** > AI-generated content. Generated by [AI System/Model Name], [Date]. Training data sources for this model include [general description — e.g., "publicly available text and image data"]. Specific source attribution is not available for this output. **CIP fields:** CIP-AI-Training-Data-Source: not-disclosed-by-vendor, CIP-AI-Generated: true. --- ## Scenario 7: Content marked AI-generated under EU AI Act Article 50 **Context:** AI-generated content that must carry machine-readable disclosure under EU AI Act Article 50, which requires providers of AI systems generating synthetic audio, image, video, or text content to mark outputs in a machine-readable format. **Attribution requirement:** Machine-readable AI-generation marking (C2PA Content Credentials recommended) plus human-visible disclosure. **Recommended language:** > This content was generated by an AI system and is marked as such in accordance with EU AI Act Article 50. AI system: [Model Name]. Provider: [Legal Entity]. Generation date: [Date]. Machine-readable provenance metadata is embedded in this file. **CIP fields:** CIP-AI-Generated: true, CIP-Content-Credentials: true. --- ## Scenario 8: Content distributed under a Provenance Certificate **Context:** AI-generated content carrying a CIP Provenance Certificate — the most complete attribution scenario, where full provenance chain is documented and verifiable through the CIP Rights Registry. **Attribution requirement:** Reference to the Provenance Certificate, which contains the complete attribution chain. **Recommended language:** > AI-generated content with full provenance record. Provenance Certificate: [Certificate ID]. Verify at [verification URL]. Source CDRs, rights composition, and generation context are documented in the certificate. **CIP fields:** CIP-Output-Provenance-Endpoint: [URL], CIP-Output-Provenance-Class: [A/B/C], CIP-AI-Output-Hash: sha256:[hex-digest]. --- ## Summary table | Scenario | Key requirement | NILP consent needed? | |----------|----------------|---------------------| | 1. Single source | Full attribution to source | No (unless identity content) | | 2. Multiple sources | Attribution to all identifiable sources | No (unless identity content) | | 3. Named style | Disclosure of imitation; no false attribution | Not required but advisable | | 4. Named voice | NILP consent documented | Yes — mandatory | | 5. Named likeness | NILP consent documented | Yes — mandatory | | 6. Untraceable derivation | Honest disclosure of limitation | No | | 7. EU AI Act Article 50 | Machine-readable + human-visible marking | Per content type | | 8. Provenance Certificate | Full provenance chain reference | Documented in certificate | --- *Creative Intellectual Property Charity — creativeip.org* *This framework is provided for educational purposes. Seek qualified legal advice for your specific circumstances.*