1. Context
The use of generative artificial intelligence tools in music production is growing at an unprecedented pace. While distribution platforms (DSPs) and organizations such as Merlin refine their policies to adapt to this reality, limbo/ takes its own stance - one rooted in our history as a company founded and run by musicians.
We believe in natural music: music that emerges from work, talent, and human expression.
Our policy on AI-generated content is not a romantic or anti-technology position - it stems from a concrete understanding of the harm it causes.
The flooding of DSPs with massive AI-generated catalogs is not a neutral phenomenon: it violates the rights of real artists whose work was used without a license to train those models, and distorts the royalty calculation mechanisms that platforms use to pay all artists. When the volume of these massive AI-generated catalogs grows disproportionately, real artists earn less.
We also know that large AI-generated catalogs are often the other side of artificial or fraudulent music consumption. They are two sides of the same problem, and limbo/ fights both.
This document sets out the criteria by which limbo/ evaluates, accepts, and distributes content, and the reasons why we ask our clients to share them.
2. The Core Rule
Content created entirely or predominantly by generative AI models trained on copyright-protected recordings without a license cannot be distributed through limbo/.
3. Evaluation Test: What Is Permitted?
The key question is whether human contribution is predominant in the final track.
Case | Assessment | Permitted |
Track 100% generated by AI (melody, arrangements, mix) | Primarily AI | ❌ No |
AI vocals over human composition, lyrics, and production | Depends on the model used | ⚠️ Conditional |
AI stem (e.g. brass) within an otherwise human-made track | Track is not primarily AI | ✅ Yes |
AI stem used as a standalone track | Primarily AI | ❌ No |
AI-assisted mastering or pitch correction | Process, not creation | ✅ Yes |
Human composition performed with virtual (non-generative) instruments | Not generative AI | ✅ Yes |
AI-generated track with a commercial license issued by the tool (exclusive or not) | Depends on the model used and the origin of its training | ⚠️ Conditional |
Practical rule: if removing the AI's contribution causes the track to cease to exist as a recognizable musical work, it is primarily AI.
4. Why Does limbo/ Not Endorse Specific Tools?
4. Why Does limbo/ Not Endorse Specific Tools?
Licensing agreements exist between organizations such as Merlin and generative platforms. However, these agreements establish frameworks for future training on licensed repertoire - they do not certify the existing model or guarantee that its outputs are exclusively the result of that training. A model partially trained on unlicensed material incorporates that learning permanently, and no technical tool currently exists that can determine what data a specific output came from.
For this reason, limbo/ does not maintain a list of approved tools. The applicable criterion is the origin of the model's training, not its commercial name - and verifying that origin is today technically impossible to guarantee.
The existence of a commercial license issued by a generative tool - whether exclusive or non-exclusive - does not equate to verification of its training origin. limbo/ accepts such content on a conditional basis, requiring the client to retain and be able to provide the license documentation. In the event of future legal or administrative action, that documentation will form part of the defense for both parties.
5. Why These Measures Protect Both Parties
5. Why These Measures Protect Both Parties
As a distributor, limbo/ is responsible before DSPs and Merlin for the content it delivers. When a release reaches the platforms, it does so under limbo/'s name and reputation.
At the same time, the risk for clients is equally concrete. The leading companies that manage recording and publishing rights globally are developing AI content detection and identification systems, combining watermarking technologies capable of identifying the generating model with automated enforcement engines. When these tools become operational - which may happen in the short term - any catalog with significant unlicensed generative AI content may be subject to analysis, claims, and potential legal action.
These policies are not an arbitrary restriction: they are the way limbo/ and its clients build catalogs that can withstand that scenario.
6. Information Required at the Time of Release Approval
6. Information Required at the Time of Release Approval
In order to properly evaluate each release, limbo/ may request that the client disclose:
Whether the track involves generative AI at any stage of its creation.
Which tool or model was used.
Whether the AI generated central elements of the track (melody, vocals, full arrangement) or peripheral elements (a stem within an otherwise human-made mix).
That the metadata — title, artist, genre, and credits — accurately reflects the actual content of the track.
limbo/ reserves the right to reject or remove from distribution any content that does not meet these conditions.
7. Other Monitored Practices
7. Other Monitored Practices
Regardless of the content's origin, limbo/ monitors and rejects:
Functional content without artistic value (white noise, nature sounds, binaural frequencies without human composition, etc.)
Metadata manipulation for algorithmic positioning (artificial SEO in titles, fake artists, incorrect genres)
Artificial or fraudulent streaming, or any practice of manipulation of play counts
8. Effective Date and Updates
8. Effective Date and Updates
This policy has been in effect since May 2026. Given the pace of market evolution and the development of new technical detection tools, limbo/ will review and update this document periodically. Clients will be notified of any substantial changes.