Music Publishers Set New AI Licensing Principles: What It Means for Songwriters
Independent music-publishing groups IMPEL and IMPF have set out a joint seven-principle framework for generative-AI licensing, with a direct demand at its centre: compositions should be valued at least equally with sound recordings. Announced on September 30, 2026, the framework addresses compensation for AI training and generated outputs, as well as the reporting and audit information publishers need to determine what their catalogues are worth in an AI deal.
The principles are not legislation, binding industry rules or a published contract. They are negotiating guardrails from two publisher bodies. But they sharpen a fault line in AI music licensing: whether the underlying song—typically administered through publishing rights—will receive a comparable share of value to the master recording when an AI company licenses music for training or commercial use.
IMPEL and IMPF put composition value at the centre
According to Music Business Worldwide’s report on the announcement, IMPEL and IMPF say compositions should be valued at least equally with sound recordings, absent other relevant factors. The groups also reject licensing models that allocate the bulk of AI revenue to master rights.
That distinction matters because recorded songs carry separate sets of rights. A master, or sound recording, concerns a particular recorded performance; composition and publishing rights concern the underlying music and lyrics. The framework is aimed at ensuring that a deal involving music does not treat the recording as the only economically meaningful input while leaving songwriters and publishers with a residual share.
The position does not prescribe a universal price, royalty split or contractual formula. Instead, it establishes the groups’ starting point for negotiations: the song must not be structurally subordinated to the recording in the allocation of AI licensing revenue.
The proposed terms reach past use, training and outputs
IMPEL and IMPF’s principles cover more than the initial act of training a model. They call for compensation related to past use, AI training, generated outputs and future exploitation. That scope is significant because an AI arrangement can raise distinct questions at different stages: material supplied to train a system, music generated through it, and the later commercial uses built around those outputs.
The framework also identifies attribution, transparency, usage reporting and deductions as core deal subjects. In practical terms, that means publishers should be able to see not merely that a licence exists, but what activity it covers and how the resulting money moves through the arrangement.
These are stated principles, rather than settled legal requirements applicable across every jurisdiction. Their value is therefore commercial and strategic: they give independent publishers a shared set of terms to press when AI companies seek access to songs, catalogues or datasets.
Transparency and limited pilots are safeguards, not side issues
For songwriters, the framework’s most consequential provisions may be its information rights. IMPEL and IMPF say rights holders must understand what is licensed, how revenue is calculated and how it is distributed. Without that visibility, a publisher may have difficulty checking whether a training licence, output-related payment or downstream revenue share has been properly accounted for.
The groups also warn that limited pilots or closed AI environments should not automatically establish broad precedents for future exploitation. That is a pointed intervention in a market where early experiments can later be cited as evidence of an accepted commercial norm. A narrow test arrangement, in the groups’ view, should not silently determine valuation or scope for a wider market.
The framework therefore connects transparency to bargaining power. Usage reporting, attribution and clarity around deductions are not simply administrative requests; they are the records that allow licensors to assess a deal, distribute revenue and challenge calculations where necessary.
Udio and KLAY point to the next licensing battleground
The joint principles arrive after the National Music Publishers’ Association announced industry-wide publisher licensing deals with Udio and KLAY on June 10, 2026. In its announcement, the NMPA said the Udio agreement provided equal compensation for publishing and recording rights in AI training. That claim is closely aligned with IMPEL and IMPF’s insistence that compositions receive at least equivalent value.
The new framework also builds on IMPF’s own earlier work. Its December 2025 AI principles called for licences covering both AI training and AI-generated exploitation, value-based compensation, transparency around datasets and usage logs, auditability, and litigation where licensing is refused.
IMPEL and IMPF’s joint intervention takes those demands into a more defined negotiating framework. The key test will be whether forthcoming AI licences deliver the visibility, auditability and composition-level compensation the groups now say should underpin the market.