Artificial intelligence is no longer knocking at the door of the music business, it is already inside the room. The recent Kobalt–ElevenLabs deal makes this reality undeniable, marking a turning point in how the industry engages with generative AI.

The deal sets a precedent by introducing a bespoke licensing framework where publishing and recordings stand on equal terms. It integrates guardrails from the outset and positions its licensing strategy as the force that will shape the market for AI-driven creativity.

Understanding the Kobalt–ElevenLabs Deal

On August 5, 2025, Kobalt announced an opt-in deal with ElevenLabs’ new AI music platform, Eleven Music. Industry watchers quickly identified it as a precedent-setting moment in generative AI licensing.

Two features make this deal stand out.

First, the economics target approximate parity between publishing and recordings, roughly a 50:50 split of royalties generated by the platform. This is a sharp break from streaming norms where publishers usually capture around 25 percent and recordings take roughly 75 percent. 

By insisting that songwriters share equally, Kobalt has reset expectations of what “fair” should mean in AI licensing.

Second is contractual protection. Kobalt secured a Most Favored Nation clause with ElevenLabs, ensuring that if any recorded music rightsholder negotiates better terms, Kobalt’s publishing terms ratchet up automatically. 

The deal also includes content protections, including third-party output detection and strict rules that prohibit naming artists, songs, labels, or publishers in their prompts. This is to ensure generated music does not infringe existing works.

ElevenLabs further confirmed that its base model is trained on licensed production music, with a Pro version in development trained on cleared catalog from Kobalt and Merlin.

This approach is striking when set against recent history. In 2024, labels sued AI startups Suno and Udio for alleged large-scale infringement. Even as those lawsuits progressed, reports suggested the majors were exploring licensing talks behind the scenes.

The Kobalt–ElevenLabs structure represents a very different posture that contrasts with the litigation-first approach.

The parity idea also departs from streaming’s historical economics, where publishing commonly lands far below masters, a dynamic tied to compulsory licensing in the United States. Since AI training is not locked into that framework, publishers can negotiate freely.

Key Takeaways From the Kobalt–ElevenLabs Deal

1. Publishing–Masters Parity

A 50:50 royalty split between compositions and recordings resets the baseline for AI licensing, countering the 25/75 streaming norm and positioning songwriters as equal stakeholders.

2. MFN as Market Stabilizer

The Most Favored Nation clause locks in fair terms by ensuring Kobalt’s deal automatically upgrades if others secure better rates. Calibrated well, it prevents a race to the bottom and accelerates broader deal-making.

3. Guardrails From the Start

Prompt restrictions, content detection, and provenance checks are embedded upfront. They curb impersonation risks and give regulators, brands, and enterprise buyers a ready-made compliance story.

How this differs from streaming economics, and why that is good for adoption

Streaming economics sit inside compulsory mechanical licensing, which constrains publisher leverage and helps explain the historical imbalance between masters and publishing. 

AI training and output licensing are not locked to that structure. That opens space for parity and for bespoke models that track product value, not just plays. This is a chance to reset incentives so both sides invest in compliant AI growth.

Unlike streaming’s one-size-fits-all model, generative AI licensing can adapt to different tiers, use cases, and revenue stacks. This makes compliance a driver of market growth rather than a drag on it.

Inside the Kobalt–ElevenLabs Deal: How Investors and Platforms Can Build a Market-Ready Licensing Framework for AI Music 

AI will not replace music as we know it, but it will reshape how rights, revenue, and reputation are managed. The question is not if, but how to build frameworks that scale. For investors, platforms, and rights holders, here are the building blocks:

(i) Training Rights: Defining What Goes In
AI models must learn from licensed music, not scraped content. Investors should back platforms that:

• Secure explicit permission from rights holders before ingestion.
• Record what works were used, when, and under what terms.
• Use tiered catalogs (e.g., production music, indie, premium, fully-cleared works) to justify different economics and controls.

This is particularly important because it gives rights holders confidence that their works aren’t misused, while making the business safer for capital.

(ii) Output Rights: Controlling What Comes Out
Not all AI outputs should be free for any use. Thus, a typical Strong framework should:

• Define approved use cases e.g., UGC, ads, games, enterprise campaigns.
• Exclude high-risk areas like artist vocal cloning or verbatim lyric copying.
• Require filters to block named-artist prompts and impersonation risks.

This is extremely important because compliance by design makes it easier for regulators, brands, and creators to trust the product. 

(iii) Economics: Making the Numbers Work
AI licensing should reflect real value and not copy streaming’s broken splits. A workable system should:

• Combine subscriptions, per-track fees, and enterprise buyouts to match customer needs.
• Adopt publishing–masters parity (50:50) as the new baseline.
• Include minimum guarantees and recoupable training fees so rights holders share in upside while investors de-risk adoption.

This way, investors get predictable returns, and creators get fairer treatment.

(iv) Contracts: MFN With Flexibility
A Most Favored Nation (MFN) clause can stop unfair undercutting, but it must be carefully scoped:

• Narrow MFN terms to specific tiers and geographies.
• Carve out space for pilots and innovation deals.

This keeps markets fair without freezing experimentation.

(v) Safety and Trust: Provenance and Audit
Traceability is non-negotiable. Frameworks should:

• Use independent fingerprinting to check outputs against catalogs.
• Embed IDs or watermarks linking outputs to model versions and datasets.
• Offer audit APIs so rights holders can check how their works are used.

This is important because transparency builds trust with brands, regulators, and creators which are necessary fuel for scaling.

(vi) Royalties and Reporting: Paying Creators on Time
In a market where royalty collection is often slow, AI licensing can set a new standard by:

Separating ledgers for training fees vs. output royalties.

• Monthly reporting and quarterly payouts to shorten cash cycles.
• Mixing CMO routing with direct pay options for enterprise clients.

This is particularly important for making Generative AI licensing investable especially for a region like Africa where transparency is a pitfall for …

(vi) Products That Create Demand
A licensing framework is only useful if it leads to products brands, creators, and agencies want. Platforms should build:

• Style packs (Afrobeats, Highlife, Gospel, Amapiano) that celebrate African sounds without impersonating artists.
• Enterprise creative seats for ad agencies, brands, and studios, bundled with compliance features.
• Creator economy tools like one-click licensing and revenue-sharing for artists who embrace AI.

Compliance should not be a barrier rather it can be repurposed as a market advantage. Demand follows when AI products are safe, clear, and culturally relevant.

Practical risk checklist for investors and companies entering the Market

1. Catalogue and rights mapping: Confirm territorial copyright ownership and splits between publishers and labels. Catalogs often have complex ownership chains. Identify works with unclear publisher assignments or orphan works and quantify remediation costs.

2. Clearance and chain of title: Prioritize fully cleared Pro-tier works for initial model training to reduce litigation risk. For other tiers, document indemnity, recoupment terms, and dispute resolution clauses.

3. Local collective societies and routing: Assess whether local collecting societies can accept and distribute payments for AI outputs. Where societies are slow or opaque, plan for direct pay mechanisms. Also, understand local copyright registration and the enforceability of assignments and moral rights.
4. Compliance and regulatory posture: Check national AI and data laws that may affect training data processing, biometric or personality rights, and consumer protection. Also, consider labelling and metadata obligations that may be adopted by regulators. Recent industry and policy work shows traceability and watermarking are likely to be part of emerging standards.

5. Commercial product fit and demand signal: Validate demand among brand buyers, agencies, game studios, and media producers for licensed style packs and enterprise seats. Run pilots with capped exposure and MGs so both sides can measure uptake.

6. Litigation exposure and insurance: Evaluate historical infringement patterns and pending global suits for precedent risk. The Suno and Udio suits show the scale of potential statutory damages for infringement in the United States. Factor potential cross-border enforcement costs into valuations.

What this means going forward

Kobalt and ElevenLabs have outlined a workable middle path. If parity, scoped MFN, opt-in governance, and robust safety become standard, AI music can exit the court-room narrative and enter a scaled, brand-safe marketplace.

We are also seeing other platforms move in this direction. Epidemic Sound’s recent launch of its Adapt tool offers a glimpse of what this future could look like. The tool allows creators to modify licensed tracks while introducing a revised compensation model for artists. This approach signals that innovation and fair value can grow together, showing how platforms can expand opportunity without undermining rights.

Labels and publishers should compete on how attractive their licensed tiers are, not on who can hold out the longest. 

Platforms should compete on compliance design and creative quality. If that happens, demand follows.

Spring Sound Ltd: Local Expertise for Global (AI) Music Licensing in Africa

At SpringSound, we help investors, platforms, and rights holders in Africa navigate the fast-changing world of music licensing. From traditional sync to emerging AI use cases, we provide the local expertise needed to unlock value in African music.

Our expertise spans music supervision, custom composition, seamless licensing, and a culturally curated library for film, TV, games, and brand advertising. We also support catalog clearance, collective rights management, and compliance with new AI and copyright laws.

We bring together legal expertise, industry know-how, and deep cultural understanding to make music licensing simple and effective. 

From helping international platforms connect with African catalogs to supporting local rights holders in unlocking new value from AI-driven creativity, Spring Sound makes complex deals clear, practical, and profitable.