Within AI Training
Who gets paid when AI trains on songs?
Training licences can pay musicians and labels, but direct deals, collective licences and opt-outs create very different winners.
On this page
- Direct catalogue deals
- Collective licensing and smaller rightsholders
- Opt in versus opt out systems
Page outline Jump by section
Introduction
The question of who gets paid when AI trains on songs is gradually shifting from a legal dispute to a market design problem. As music companies, publishers, collecting societies and AI developers explore licensing arrangements, the central challenge is not simply whether training should be licensed, but how payments should be distributed across labels, songwriters, performers and independent creators. The answer varies dramatically depending on the licensing model. Some systems favour large catalogue owners that can negotiate directly with AI companies. Others aim to spread revenue more broadly through collective management. Still others focus on giving creators a choice through opt-in or opt-out participation. The structure chosen could determine whether AI training becomes a new income stream for the wider music industry or primarily benefits those with the largest rights portfolios. [U.S. Copyright Office]copyright.govCopyright OfficePart 3: Generative AI Training pre-publication versionMay 6, 2025 — This section describes how and why copyrighted works…
Direct Catalogue Deals
The fastest-emerging model is the direct licensing agreement between AI developers and major music rights holders. Under this approach, an AI company negotiates access to specific catalogues and pays for the right to use recordings, compositions or both during model training.
This model appeals to large rights owners because it resembles existing music licensing practices. Major record companies already control extensive catalogues, have licensing teams and can negotiate bespoke commercial terms. Recent negotiations and settlements involving AI music companies such as Suno and Udio have increasingly centred on licensed access to catalogue content rather than purely theoretical fair-use arguments. Reports indicate that major labels have pursued agreements allowing AI training in exchange for compensation and platform controls. Forbes [reuters]reuters.comThis agreement follows a similar settlement between Warner Music, Universal Music Group (UMG), and another AI platform, Udio. These settl… The advantages are straightforward:
- Rights holders know exactly who is using their music.
- AI developers obtain legal certainty.
- Payment terms can be customised.
- Catalogue owners can negotiate transparency, attribution and usage restrictions.
The weakness is concentration. Direct licensing naturally favours companies with large repertoires and strong bargaining power. A major label controlling millions of recordings can negotiate from a position that most independent artists cannot match. Smaller rights holders may find themselves excluded from negotiations altogether or offered less favourable terms. This creates a risk that AI licensing revenue becomes concentrated among the largest catalogue owners while much of the long tail of music receives little compensation. Forbes [pitchfork]pitchfork.comKlay utilizes its Large Music Model, trained exclusively on licensed content, to offer interactive, human-centered music experiences and… Another unresolved issue is internal distribution. Even if a label receives substantial licensing revenue, questions remain about how much reaches featured artists, session musicians, songwriters and producers. Existing recording contracts often were not written with AI training uses in mind, creating uncertainty about entitlement to future payments. [councilmusicmakers.org]councilmusicmakers.orgThe music-maker perspective on the music industry's AI deals6 Nov 2025 — Explicit Consent: Music-makers must grant explicit consent befor…
Can Collective Licensing Solve the Scale Problem?
Direct deals work best when rights ownership is concentrated. Music rights, however, are often fragmented across multiple parties. A single song may involve several songwriters, publishers, performers and recording owners. For AI developers seeking large-scale datasets, negotiating with every individual rights holder is impractical.
Collective licensing attempts to solve this problem by allowing a single organisation to grant licences on behalf of many members and distribute revenue afterwards. Music already uses collective management extensively for public performance and other rights. Some policymakers and industry groups have suggested similar mechanisms for AI training. The idea is that AI developers would pay into a licensing scheme, while collecting societies or designated bodies would allocate revenue among participating rights holders. [GOV.UK]GOV.UKreport on copyright and artificial intelligence18 Mar 2026 — Report on Copyright and Artificial Intelligence. Presented to Parliament pursuant to Section 136 of the Data (Use and Acces…
The attraction is efficiency. Instead of thousands of negotiations, an AI company could obtain broad access through a single licence. Smaller creators would also gain access to a market they might otherwise struggle to enter. In principle, collective licensing could reduce transaction costs while expanding participation. [OUP Academic]academic.oup.comOUP AcademicCopyright and AI training data—transparency to the rescue?by A Buick · 2025 · Cited by 90 — AI developers to be required by l…
Real-world experimentation is beginning to appear. Sweden’s collecting society STIM introduced an AI-related licensing framework intended to allow authorised use of songs while generating compensation for songwriters, illustrating how collective approaches might operate at scale. [plasticheadmedia.com]plasticheadmedia.comAI and Training Data Licensing: What It Means for…May 7, 2026 — In September 2025, STIM (the Swedish Performing Rights Society) rolled…
Yet collective systems raise difficult implementation questions:
- How should revenue be divided among millions of works?
- Should payment depend on catalogue size, training frequency or measurable influence on model outputs?
- How are independent artists represented?
- What happens when rights ownership is disputed or incomplete?
These questions are not merely administrative. They determine whether collective licensing becomes a genuinely inclusive system or simply another mechanism that favours the largest repertoires.
Why Attribution Matters for Payment
One reason licensing remains difficult is that AI training does not resemble traditional music consumption. A streaming service can count plays. A radio station can track broadcasts. AI models learn from vast numbers of recordings simultaneously.
As a result, many proposed payment systems struggle to identify whose music contributed most to a model’s capabilities. Researchers have begun developing methods for training-data attribution, attempting to measure which works influenced particular outputs or model behaviour. Such systems could eventually support more granular compensation schemes. [arXiv]arxiv.orgarXivLarge-Scale Training Data Attribution for Music Generative Models via UnlearningJune 23, 2025…
However, attribution at music-industry scale remains technically challenging. Current models typically learn patterns from enormous datasets rather than drawing directly from identifiable songs. This makes it difficult to calculate royalty-style payments based on individual contributions. Consequently, many licensing proposals rely on broader allocation formulas rather than precise measurement of influence. [arXiv]arxiv.orgarXivLarge-Scale Training Data Attribution for Music Generative Models via UnlearningJune 23, 2025…
The attribution problem helps explain why transparency has become such a significant policy issue. Rights holders increasingly argue that developers should disclose training materials so creators can verify participation and claim compensation where appropriate. Transparency requirements are also seen as a prerequisite for meaningful licensing markets. [OUP Academic]academic.oup.comOUP AcademicCopyright and AI training data—transparency to the rescue?by A Buick · 2025 · Cited by 90 — AI developers to be required by l…
Opt-In Versus Opt-Out Systems
Perhaps the most contested implementation choice is whether participation should be opt-in or opt-out.
An opt-in system requires affirmative permission before music can be included in a training dataset. Supporters argue that this approach best respects copyright and creator autonomy. Artists, labels and publishers decide whether to participate and under what conditions. Several emerging licensed AI initiatives have emphasised creator consent and voluntary participation. [councilmusicmakers.org]councilmusicmakers.orgThe music-maker perspective on the music industry's AI deals6 Nov 2025 — Explicit Consent: Music-makers must grant explicit consent befor… [MusicRadar The advantages include:]musicradar.comThis deal follows a recent similar agreement between Warner and Udio, and is part of the music industry's broader strategy to combat the…
- Clear permission records.
- Strong creator control.
- Easier enforcement of contractual terms.
- Greater confidence that compensation reflects consent.
The downside is scale. Building comprehensive datasets becomes more difficult when every rights holder must actively join. Smaller creators may never hear about licensing opportunities, and incomplete participation can reduce dataset diversity.
Opt-out systems take the opposite approach. Music can be included unless a rights holder explicitly withdraws permission. Advocates argue that this reduces friction and enables broader licensing coverage. Critics counter that many creators may not realise their works are being used or may lack the resources to monitor participation effectively. [OUP Academic]academic.oup.comOUP AcademicCopyright and AI training data—transparency to the rescue?by A Buick · 2025 · Cited by 90 — AI developers to be required by l…
The choice between opt-in and opt-out therefore reflects a deeper policy trade-off. Opt-in maximises control but may limit scale. Opt-out maximises coverage but risks weakening meaningful consent. The debate increasingly centres on whether efficiency gains justify shifting responsibility from AI developers to creators. [GOV.UK]GOV.UKreport on copyright and artificial intelligence18 Mar 2026 — Report on Copyright and Artificial Intelligence. Presented to Parliament pursuant to Section 136 of the Data (Use and Acces…
The Emerging Battle Over Value Distribution
As lawsuits, settlements and licensing negotiations evolve, the music industry is moving towards a licensing-first environment rather than relying entirely on unresolved fair-use arguments. Governments, industry bodies and rights organisations increasingly discuss frameworks built around permission, transparency and compensation. [UK Music]ukmusic.orgUK MusicUK Music Chief Hails House of Lords Report Highlighting…3 days ago — The Government's task should be to create the conditions… [GOV.UK]GOV.UKreport on copyright and artificial intelligence18 Mar 2026 — Report on Copyright and Artificial Intelligence. Presented to Parliament pursuant to Section 136 of the Data (Use and Acces…
The most important unresolved question is not whether licensing markets can exist. Evidence suggests they already are emerging. The harder question is how revenue will be distributed once those markets mature. Direct deals reward negotiating power. Collective licences promise broader participation but require difficult allocation rules. Opt-in systems prioritise creator choice, while opt-out systems prioritise scale. Forbes [pitchfork]pitchfork.comKlay utilizes its Large Music Model, trained exclusively on licensed content, to offer interactive, human-centered music experiences and… Who gets paid when AI trains on songs will ultimately depend less on the technology itself than on which of these licensing models becomes dominant. The design choices made now will shape whether AI training creates a narrow revenue stream for large catalogue owners or a broader market that includes the musicians, songwriters and independent creators whose work helped teach the systems in the first place. [councilmusicmakers.org]councilmusicmakers.orgThe music-maker perspective on the music industry's AI deals6 Nov 2025 — Explicit Consent: Music-makers must grant explicit consent befor… [UK Music]ukmusic.orgUK MusicUK Music Chief Hails House of Lords Report Highlighting…3 days ago — The Government's task should be to create the conditions…
Amazon book picks
Further Reading
Books and field guides related to Who gets paid when AI trains on songs?. Use these as the next step if you want deeper reading beyond the article.
All You Need to Know About the Music Business
Explains royalties, licensing structures, rights ownership and payment flows that underpin AI-training compensation debates.
The Future of the Professions
Provides context for how AI reshapes compensation systems and professional value creation.
The Coming Wave
Offers accessible background on AI governance, incentives and regulation relevant to training-licensing markets.
Copyright in the Music Industry
First published 2021. Subjects: Written communication.
eBay marketplace picks
Marketplace Samples
Example marketplace items related to this page. Use the search link to explore similar finds on eBay.
Endnotes
-
Source: copyright.gov
Link: https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdfSource snippet
Copyright OfficePart 3: Generative AI Training pre-publication versionMay 6, 2025 — This section describes how and why copyrighted works...
Published: May 6, 2025
-
Source: GOV.UK
Title: report on copyright and artificial intelligence
Link: https://www.gov.uk/government/publications/report-and-impact-assessment-on-copyright-and-artificial-intelligence/report-on-copyright-and-artificial-intelligenceSource snippet
18 Mar 2026 — Report on Copyright and Artificial Intelligence. Presented to Parliament pursuant to Section 136 of the Data (Use and Acces...
-
Source: forbes.com
Link: https://www.forbes.com/sites/virginieberger/2025/06/06/what-suno-and-udios-ai-licensing-deals-with-music-majors-could-mean-for-creators-rights/Source snippet
ForbesWhat Suno And Udio's AI Licensing Deals With Music...Jun 6, 2025 — The same companies are at the table hammering out AI music lice...
-
Source: reuters.com
Link: https://www.reuters.com/legal/litigation/warner-music-group-settles-copyright-case-with-suno-licensed-ai-music-2025-11-25/Source snippet
This agreement follows a similar settlement between Warner Music, Universal Music Group (UMG), and another AI platform, Udio. These settl...
-
Source: pitchfork.com
Link: https://pitchfork.com/news/warner-music-group-signs-licensing-deal-with-ai-music-company-klaySource snippet
Klay utilizes its Large Music Model, trained exclusively on licensed content, to offer interactive, human-centered music experiences and...
-
Source: councilmusicmakers.org
Link: https://councilmusicmakers.org/musicai-nov2025/Source snippet
The music-maker perspective on the music industry's AI deals6 Nov 2025 — Explicit Consent: Music-makers must grant explicit consent befor...
-
Source: academic.oup.com
Link: https://academic.oup.com/jiplp/article/20/3/182/7922541Source snippet
OUP AcademicCopyright and AI training data—transparency to the rescue?by A Buick · 2025 · Cited by 90 — AI developers to be required by l...
-
Source: plasticheadmedia.com
Link: https://plasticheadmedia.com/ai-and-training-data-licensing-what-it-means-for-musicians-today/Source snippet
AI and Training Data Licensing: What It Means for...May 7, 2026 — In September 2025, STIM (the Swedish Performing Rights Society) rolled...
Published: May 7, 2026
-
Source: arxiv.org
Link: https://arxiv.org/abs/2506.18312Source snippet
arXivLarge-Scale Training Data Attribution for Music Generative Models via UnlearningJune 23, 2025...
Published: June 23, 2025
-
Source: musicradar.com
Link: https://www.musicradar.com/music-tech/artists-and-songwriters-will-have-full-control-over-whether-and-how-their-names-images-likenesses-voices-and-compositions-are-used-in-new-ai-generated-music-warners-and-suno-link-up-to-create-new-legal-ai-platformsSource snippet
This deal follows a recent similar agreement between Warner and Udio, and is part of the music industry's broader strategy to combat the...
-
Source: forbes.com
Link: https://www.forbes.com/sites/virginieberger/2025/12/18/launch-train-settle-how-suno-and-udios-licensing-deals-made-copyright-infringement-profitable/Source snippet
Launch, Train, Settle: How Suno And Udio's Licensing...18 Dec 2025 — Suno and Udio operated for approximately two years, training their...
-
Source: forbes.com
Link: https://www.forbes.com/sites/virginieberger/2025/12/29/nine-predictions-for-the-music-industry-in-2026-how-ai-reshapes-licensing-and-power/Source snippet
Nine Predictions For The Music Industry In 2026: How AI...29 Dec 2025 — Nine predictions for how AI will reshape music licensing, catalo...
-
Source: copyright.com
Title: CC C Launching New AI Content Re-Use Rights for U.S
Link: https://www.copyright.com/media-press-releases/ccc-launching-new-ai-content-re-use-rights-for-u-s-academic-customers-and-transactional-licensing-capabilities-for-ai/Source snippet
Mar 3, 2026 — The Organization's Licensing Portfolio Will Feature Four AI Licensing Options to Support Internal and External AI Use Cases...
-
Source: ukmusic.org
Link: https://www.ukmusic.org/news/uk-music-chief-hails-house-of-lords-report-highlighting-clear-and-present-danger-of-ai-training/
Additional References
-
Source: landr.com
Link: https://www.landr.com/fairaiSource snippet
LANDR Fair Trade AI ProgramUsers distributing music with LANDR can opt-in to the program to earn money by allowing their content to be us...
-
Source: reddit.com
Link: https://www.reddit.com/r/SunoAI/comments/1olqy3o/udio_vs_suno_implications_license_distributions/Source snippet
UDIO vs SUNO IMPLICATIONS. LICENSE...For those who may not be aware Udio settled the lawsuit against them for training their AI with com...
-
Source: musicbusinessworldwide.com
Link: https://www.musicbusinessworldwide.com/umg-and-sony-seek-to-add-61000-copyrighted-works-to-suno-lawsuit-after-discovery-reveals-suno-trained-on-millions-of-their-recordings/Source snippet
UMG and Sony seek to add over 61k recordings to Suno...5 days ago — Universal Music Group and Sony Music Entertainment have asked a fede...
-
Source: wsj.com
Link: https://www.wsj.com/business/media/universal-music-settles-with-ai-firm-udio-c926ca39Source snippet
As part of the agreement, Udio will launch a licensed subscription-based music creation platform in 2026, allowing users to remix and cus...
-
Source: riaa.com
Link: https://www.riaa.com/record-companies-bring-landmark-cases-for-responsible-ai-againstsuno-and-udio-in-boston-and-new-york-federal-courts-respectively/Source snippet
Record Companies Bring Landmark Cases for...24 Jun 2024 — Unlicensed services like Suno and Udio that claim it's 'fair' to copy an artis...
-
Source: musicbusinessworldwide.com
Link: https://www.musicbusinessworldwide.com/suno-moves-to-keep-size-of-its-ai-training-data-sealed-in-umg-and-sonys-copyright-case-citing-competitive-harm/Source snippet
Suno moves to keep size of its AI training data sealed in...1 day ago — Suno moves to keep size of its AI training data sealed in UMG an...
-
Source: prsformusic.com
Link: https://www.prsformusic.com/works/how-copyright-works/ai-and-music-copyright -
Source: musicbusinessworldwide.com
Link: https://www.musicbusinessworldwide.com/universal-warner-and-sony-in-talks-to-license-ai-music-generators-suno-and-udio-report/Source snippet
Labels in licensing talks with AI music generators Suno...2 Jun 2025 — The major music companies are reportedly in licensing talks with...
-
Source: linkedin.com
Link: https://www.linkedin.com/posts/michael-campbell-linn_this-is-what-i-believe-will-happen-in-music-activity-7422579263641391104-yTZ3Source snippet
How AI Will Change Copyright Regulations · Challenges in Music Licensing for 2025 · How AI Influences Music Licensing...Read more...
-
Source: sonymusic.com
Title: klay signs ai licensing deals universal sony music [publishing]({{ ‘publishing/’ | relative_url }}) warner chappell
Link: https://www.sonymusic.com/sonymusic/klay-signs-ai-licensing-deals-universal-sony-music-publishing-warner-chappell/Source snippet
Music Technology Company Klay Signs First-Of-Its-Kind AI...20 Nov 2025 — Music Technology Company Klay Signs First-Of-Its-Kind AI Licens...
Topic Tree