How music rights work
Every piece of recorded music carries two separate copyrights.
The recording (master). The copyright in a specific recorded performance. It is owned by whoever paid for and produced the recording: usually a record label, or the artist in the case of independent releases. Using a recording in a product, video or dataset requires a license from the recording owner.
The composition. The copyright in the underlying musical work: melody, harmony and lyrics. It is owned by the songwriters and administered by their publishers. Using any recording of a song also requires a license from the composition owners.
A single song can therefore need two licenses from two sets of rights holders, and each side is often split across several owners. Both sides must be cleared for a use to be licensed.
License types
| License type | What it covers |
|---|---|
| Sync license | Music used alongside visual media. Traditionally negotiated per placement. |
| Streaming license | Music in consumer audio or video streaming. Typically administered through collecting societies. |
| Mechanical license | Reproduction of a composition, for example in a download. |
| AI license | AI uses of music, priced by use-case risk: search, classification and recommendation at the low end, generative output at the high end. |
| Generative AI license with downstream rights | Training or tuning a model on music, plus commercial use of what that model generates. |
| All-media license | One track across all media and platforms. The Rights Foundry’s standard commercial license. |
What changes in an AI context
Training rights. Using music as data to train a model. Most existing commercial music licenses do not cover this use.
Fine-tuning rights. Using a subset of music to adjust a trained model toward a style, genre or use case. Outputs may carry identifiable traits of the licensed music, which raises the stakes.
Evaluation rights. Using music in fixed sets to benchmark classification, recommendation or generation quality.
Generative output rights. Whether music a model generates can be used commercially. This is the right a commercial product actually needs. The Rights Foundry builds downstream commercial rights into its generative AI license explicitly.
Why this is changing now
The market moved to licensed inputs. More than 100 AI copyright cases have been filed since 2023 (Information is Beautiful, AI copyright tracker, June 2026). Major rights holders have moved from suing AI music companies to licensing them. Those deals are private: each covers its own catalog and stops at its own walls.
The courts weighed in. In July 2026 a Munich court ruled in GEMA v. Suno that training a music generation model on unlicensed recordings infringes, and that users in the EU bring EU law wherever the model is built.
Regulators want evidence. Under the EU AI Act, providers of general-purpose AI models must publish a summary of the content used for training and keep a copyright policy that honors machine-readable opt-outs. The European Commission began enforcing these obligations on 2 August 2026.
Audio became a product requirement. Leading video generation models now produce audio natively. Every app, game and creator tool needs music it can ship.
Rights before use, not policed after
Fingerprinting and enforcement systems such as YouTube Content ID detect music after it has been used, inside their own platforms. The Rights Foundry works the other way round. It confirms what is permitted before any use happens, records the use as it happens, and pays the rights holders from it.
Where The Rights Foundry sits
The Rights Foundry is AI rights infrastructure: the permission layer for AI and creative IP, starting with music. It serves three groups through the same catalog, the same rights architecture and the same API:
| Group | What they use it for |
|---|---|
| Builders | Self-serve API and MCP server for apps, games, video and creator tools. Free to start, usage-based pricing. |
| Brands and platforms | Enterprise plans for cleared music at volume across campaigns, creator programs and products, on one agreement. |
| AI and data teams | Rights-cleared training, fine-tuning and evaluation data, with consent records, rights flags and downstream commercial rights. |
Traditional sync licensing remains the right model for film, television and major campaigns. The Rights Foundry serves the uses that model cannot reach: product speed, machine scale, and legal certainty at the point of integration.
Foundation
The Rights Foundry launches with Audiosocket as its first catalog provider:
- 110,000+ globally cleared songs
- 35,000+ tracks cleared for AI under signed GenAI Addenda
- 3,500+ independent artists
- Recording and composition cleared together