Key terms in this module(10)
Quick definitions for the terms used here. Open any term for the full entry.
- Biometric rights
Rights over a person's biometric identifiers (face, voice, gait, fingerprints, distinctive movement patterns). UK GDPR Article 9 special category data; EU GDPR equivalents; jurisdiction-specific personality-rights frameworks elsewhere.
- cip.md
The framework's declaration file format. A plain-text Markdown file at https://[domain]/.well-known/cip.md declaring rights, consent, and certification status in machine-readable form. Specified in the v3.18 consolidated cip.md Specification page; AI Generation field set added v3.24.
- Distribution
Dissemination, sale, licensing, or transfer of copies to the public. Includes movement of content through APIs, feeds, search, recommendation systems, syndication. Foundational rights category.
- Inference
Real-time computation of outputs from inputs using learned parameters. The operational stage where AI model meets specific input to produce a specific output.
- Neighbouring rights
Rights held by performers, phonogram producers, and broadcasters in creative works, distinct from copyright. Subsist independently; not waived by AI ingestion.
- 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.
- NILP Downstream Obligation
The framework's position that commercial deployers of AI-generated identity content carry the rights-holder claim where the deployment proceeds without explicit consent verification. The obligation flows downstream from training-data ingestion through output generation to commercial deployment.
- Output
Generated content instance, text, image, audio, action. The single-instance unit produced by inference.
- Platform
System that hosts models + applies governance, policies, ranking logic, and monetisation rules. Distinct from Model Provider (which makes the model) and Host (which serves it).
- 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.
4.1 What NILP covers
Name rights — the right to control commercial use of your name. Includes use in AI-generated advertising, endorsements, or content that implies your association with a product or service without consent.
Image rights — the right to control use of your likeness. AI-generated images of real people engage image rights even where no actual photograph was used.
Voice rights — increasingly recognised as a distinct category. Voice is the most commercially replicable biometric identifier. A voice clone made from recordings engages voice rights, neighbouring rights, and biometric rights simultaneously.
Publicity rights — the right to control the commercial value of your identity. Particularly strong in some US state jurisdictions.
4.2 How AI engages NILP rights — the four pipeline stages
Training data ingestion — AI systems training on images, audio recordings, or biographical data of a specific person. Requires consent from the person, not from any platform that hosted the content.
Fine-tuning — voice model fine-tuning is the most direct NILP engagement in AI. A model fine-tuned specifically to replicate a specific artist's voice requires a NILP Licence.
Inference — generated outputs — when an AI system generates identity content without authorisation, it creates NILP Downstream Obligation liability.
Distribution and commercial use — when AI-generated identity content is published or used commercially, the rights holder has a claim regardless of how many intermediary steps occurred.
4.3 The NILP pipeline licence flow
A NILP Downstream Obligation is the liability that attaches when AI-generated identity content is used commercially without a valid underlying NILP Licence. It travels with the output and can be asserted by the rights holder regardless of intermediaries.
4.4 Minimum NILP protection declaration in cip.md
The seven fields required for minimum NILP protection:
- CIP-NILP-Protected: true
- CIP-NILP-Deepfake: Prohibited
- CIP-NILP-Voice-Clone: Prohibited
- CIP-NILP-Likeness-AI: Prohibited
- CIP-NILP-Commercial-Use: Prohibited-Without-Licence
- CIP-Biometric-Training: Prohibited
- CIP-NILP-Licence-Contact: rights@yourdomain.com
Summary
Key Takeaways
- NILP rights cover name, image, likeness, voice, and publicity
- AI engages NILP at training, fine-tuning, inference, and distribution stages
- The NILP Downstream Obligation travels with AI-generated identity content
- Performers hold both NILP rights and neighbouring rights — both engaged by voice cloning
- The seven-field cip.md NILP declaration is the minimum required protection
Self-check
Check Your Understanding
- What does the acronym NILP stand for in the CIP framework?
- At which AI pipeline stage is voice cloning the most direct NILP engagement?
- What is a NILP Downstream Obligation?
- A voice clone of a performer is created from their recordings. Which rights are simultaneously engaged?
- How many fields does the minimum NILP protection declaration in cip.md require?