Generative AI is rewriting the rules of the music ecosystem, at speed and scale we have never seen before. For independent publishers and the songwriters and composers they represent, this shift brings both opportunity and profound risk. At the core lies a simple truth: AI systems cannot exist without the creative works that fuel their training and outputs. Yet without clear licensing frameworks, rightsholders risk being excluded from the economic value their works generate.

IMPF is urging local CMOs to take decisive steps to safeguard the future of music and ensure human creativity is not left behind. As key partners for independent music publishers, CMOs are central to building fair and sustainable solutions, and IMPF stands ready to support them in this mission.

In this context, the IMPF Key Principles have been developed to serve as a foundation for dialogue. At the heart of these principles is a dualpillar licensing model, covering both training and exploitation, that guarantees fair compensationat every stage of AI’s use of music.

Furthermore, CMOs should be prepared to pursue litigation, not as a last resort, but as a vital tool to establish precedent and compel fair licensing. 

This is not only about protecting revenue streams. It is about defending the integrity of human creativity and ensuring that the music ecosystem evolves on fair and sustainable terms, with CMOs leading the way.

IMPF Key Principles for CMO Licensing Models for Generative AI

1. Dual-Pillar Compensation Structure

1. Pillar 1: Training-Based Licensing

1. Scope of Licensing

• Covers all generative AI providers operating in the jurisdiction, regardless of where or when the training occurred.
• Applies to any use of protected musical works during model training, including: 
• Pre-training on large datasets
• Fine-tuning with curated music catalogues. 
• Use of synthetic data derived from copyrighted works

1. Compensation Structure

• There should be at least parity in compensation between the sound recording and the underlying musical composition. 
• Includes a minimum royalty guarantee, ensuring baseline compensation even if income is low or delayed.
• Designed to reflect the ongoing value extraction from training data, not just a one-time fee.
• Usage-based licensing (per work used, per subscription…).
• Collective licensing schemes through PROs/CMOs for efficiency.

1. Value Attribution

• Recognises that the success of AI models is directly tied to the quality and diversity of the training data, which often includes copyrighted music.
• Treats training as a commercial exploitation, not merely technical input.

1. Enforcement & Transparency

• Requires AI providers to disclose training datasets and usage logs.
• CMOs may implement auditing mechanisms to verify compliance.
• Promotes standardised metadata tagging to trace content lineage and ensure fair attribution.

1. Pillar 2: Exploitation-Based Licensing

1. What It Covers

• Downstream uses of music generated by AI systems trained on copyrighted works: 
• Streaming platforms (Spotify, YouTube, etc.)
• Public performance (bars, gyms, retail stores)
• Background music in apps, websites, and smart devices
• Commercial music tools that allow users to generate custom tracks

1. Compensation Mechanism

• Rightsholders are entitled to a share of all economic benefits derived from AI-generated music.
• This includes: 
• Revenue from subscriptions or licensing fees paid by end users
• Advertising revenue tied to AI-generated content
• Royalties from performance or reproduction of AI music
• The model ensures that rightsholders are compensated even if their original works aren’t directly identifiable in the output.

1. Attribution & Fairness

• Recognises that AI-generated music is only possible because of the original creative works used in training.
• Treats AI outputs as derivative works, even if they are not exact copies.

1. Enforcement & Monitoring

• Platforms using AI-generated music must: 
• Report usage data to CMOs
• Pay royalties based on usage volume and revenue
• Tag AI-generated content to distinguish it from human-created works

3. Value-Based Royalty Allocation

• Royalties must be based on actual market value generated by AI content.
• Recognise that AI outputs derive their value from original human-created works.
• Include synthetic data in scope, as it often originates from human creativity.

4. Transparency & Auditability

• Require full disclosure of training datasets and usage logs.
• Implement mechanisms for CMOs to audit AI providers and platforms.
• Encourage standardised metadata tagging to trace content lineage.

5. Market Inclusivity

• Apply licensing obligations to all AI providers operating in the market, regardless of where training occurred.
• Ensure small and large providers alike are subject to equitable licensing terms.

6. Continuous Adaptation

• Include review clauses and feedback loops with stakeholders (creators, platforms, developers).

7. Moral Rights Protection

• Prevent outputs that distort, demean, or misrepresent an creator’s work or reputation.

II. Litigation 

Consider initiating litigation in order to push for adoption of a licensing framework for GenAI