Close-up of a synthesizer keyboard's control panel — knobs, sliders, and piano keys
Copyright · AI Training Data
Photo credit: Skitter Photo / stocksnap · CC0 1.0

Your Music and Art in AI Training Data: What Creators Need to Know

TL;DR

AI companies have trained models on songs, images, and written works, and whether that is lawful is being decided in courts. In May 2025, the U.S. Copyright Office concluded that some uses of copyrighted works for generative AI training will qualify as fair use and some will not. Major labels have already moved to licensing deals. Independent creators still have options, and knowing them matters.

Aghil Ebrahimi, Esq.
Licensed in California · Ontario · Quebec~9 min read

The AI tools generating music, images, and written content today were trained on millions of works created by human artists, musicians, photographers, writers, and developers.

Whether that is legal is now being litigated in U.S. courts. What you can do about it follows from the answer.

Here is where things stand.

How AI training works and why your work is in it

Generative AI systems learn by processing enormous amounts of existing content. A music generation model is trained on recordings, compositions, and audio files. An image model is trained on photographs, illustrations, and digital art. A text model is trained on books, articles, websites, and code. The model extracts patterns from this material, then uses those patterns to generate new outputs on demand.

According to its developers, LAION-5B, an image-text dataset released in 2022, consists of 5.85 billion image-text pairs drawn from Common Crawl web data.

According to the Copyright Office, some AI companies have licensed works for training, while others have relied on fair use.

That position is now being tested in court.

On May 9, 2025, the U.S. Copyright Office released a pre-publication version of Part 3 of its Report on Copyright and Artificial Intelligence. Part 3 addresses the specific question of AI training data.

The Office's conclusion on fair use: "some uses of copyrighted works for generative AI training will qualify as fair use, and some will not." The Office said the result will depend on what works were used, from what source, for what purpose, and with what controls on the outputs, and that commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially through illegal access, "goes beyond established fair use boundaries."

That is not a court ruling. It is the Copyright Office's legal analysis. It does not resolve the lawsuits currently in progress.

On licensing, the Office concluded that government intervention would be premature at this time and recommended allowing the licensing market to continue to develop; where market failures are shown for specific types of works in specific contexts, it said targeted intervention such as extended collective licensing should be considered.

What the courts have actually decided

Thomson Reuters v. Ross Intelligence

On February 11, 2025, the federal district court in Delaware ruled on summary judgment in Thomson Reuters v. Ross Intelligence that Ross's use of Westlaw headnotes to build a competing AI legal-research tool was not fair use. The court found factors one and four favored Thomson Reuters and factors two and three favored Ross, and it noted that only non-generative AI was before it.

On September 29, 2026, the U.S. Court of Appeals for the Third Circuit affirmed, holding that the headnotes are copyrightable and that Ross's use was not fair use (No. 25-2153).

The music industry lawsuits: Suno and Udio

On June 24, 2024, record companies filed copyright infringement lawsuits against Suno and Udio, two AI music generation platforms. Infringement means using a copyrighted work in a way the law reserves to its owner, without permission (17 U.S.C. § 501(a)). The plaintiffs included UMG Recordings, Inc., Sony Music Entertainment and Warner Music Group labels. The lawsuits alleged that the platforms copied copyrighted sound recordings, without permission, to train their AI models.

Two record companies, Universal Music Group (UMG) and Warner Music Group (WMG), have announced three settlements in these cases. After each settlement, the settling company dropped its own claims (a voluntary dismissal), and the other labels' claims continued. No court ruled on fair use in reaching these settlements.

UMG and Udio announced a settlement on October 29, 2025. In its announcement that day, UMG said the deal includes a compensatory legal settlement and new licenses for recorded music and publishing. UMG also said that Udio's new platform would launch in 2026 and would be trained on authorized and licensed music (UMG announcement). The court approved UMG's dismissal on November 5, 2025.

WMG and Udio announced a settlement on November 19, 2025. WMG said that day that Udio's licensed AI music creation service was set to launch in 2026. WMG said the service would use AI models trained on licensed and authorized music (WMG announcement). The court approved Warner's dismissal on November 25, 2025.

WMG and Suno announced a partnership on November 25, 2025. In WMG's words, the deal "settles previous litigation between the companies." WMG said that day that Suno would launch new licensed models in 2026 and retire its current models (WMG announcement). Warner's labels filed a stipulation dismissing their claims against Suno on December 9, 2025.

As of November 10, 2026, Sony is still suing Udio, and UMG and Sony are still suing Suno:

Sony v. Udio. Sony's claims in the 2024 case are pending in the Southern District of New York (UMG Recordings, Inc. v. Uncharted Labs, Inc., No. 1:24-cv-04777). On June 29, 2026, the court denied Sony's request to amend (change) its complaint, the document that starts a lawsuit. On July 20, 2026, Sony and affiliated labels filed a second lawsuit against Udio in the same court (Sony Music Entertainment v. Uncharted Labs, Inc., No. 1:26-cv-06120).

UMG and Sony v. Suno. The 2024 case is pending in the District of Massachusetts (UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611). UMG, Capitol Records and Sony filed an amended complaint on August 25, 2026. The court's October 30, 2026 scheduling order sets June 22, 2027 as the deadline for dispositive motions (requests to decide all or part of the case without a trial). On September 18, 2026, UMG, Sony and affiliated labels filed a second lawsuit against Suno in the same court (UMG Recordings, Inc. v. Suno, Inc., No. 1:26-cv-14275). On October 30, 2026, Suno asked the court to pause that second case (a motion to stay).

None of these four cases has reached a final judgment.

What this means if you are an independent creator

The settlements above are licensing agreements between record companies and AI platforms. According to the labels' October and November 2025 announcements, the platforms' new models would use licensed music training data. UMG said on October 29, 2025 that its new licenses will provide further revenue opportunities for UMG artists and songwriters. WMG said on November 25, 2025 that its artists and songwriters will have full control over the use of their work in new AI-generated music. That control covers their names, images, likenesses, voices and compositions. WMG also described benefits for the wider creative community and for people using Suno to make music.

Independent creators in this landscape face three realities:

Public web content has been used to build AI training datasets. Some training datasets were built from public web content. According to its developers, LAION-5B, described above, was drawn from Common Crawl web data. We recommend assuming that music, photos or art you have posted publicly is in a training dataset. The FAQ below explains how to investigate a specific model.

The date you registered limits what you can recover for past training. Statutory damages and attorney's fees are not available for an infringement that began before your registration took effect (17 U.S.C. § 412). The FAQ "Can I sue an AI company for training on my work without permission?" below explains the exceptions to that rule. It also explains what you can still recover. Before you sue alone, we recommend asking a copyright attorney whether a class action fits your situation.

Courts have reached different results so far. In Thomson Reuters v. Ross (D. Del. Feb. 11, 2025), the court rejected fair use for a non-generative AI tool, and the Third Circuit affirmed on September 29, 2026. In Bartz v. Anthropic (N.D. Cal. June 23, 2025), the court granted summary judgment that training was fair use but did not extend that ruling to pirated library copies; on July 20, 2026, it granted final approval of a $1.5 billion class settlement. In Kadrey v. Meta (N.D. Cal. June 25, 2025), the court granted Meta summary judgment on its fair use defense on that record, and stated that the ruling does not stand for the proposition that Meta's use is lawful. Bartz and Kadrey are district-court rulings.

What you can do now

We recommend the following steps.

Register your copyright. Registration does not retroactively fix past unlicensed training. For a United States work, you must register or preregister before you can sue for infringement, subject to the exceptions in 17 U.S.C. § 411(a) and (c). Statutory damages and attorney's fees are available only if the work was registered before the infringement began or, for an infringement that began after first publication, within three months after that publication, subject to the exceptions in 17 U.S.C. § 412 (see the FAQ below). We recommend registering your catalog.

Use available opt-out mechanisms. Opt-out controls for training data use exist only where an AI company or platform offers them, on that provider's terms. The U.S. Copyright Office's May 2025 pre-publication report on generative AI training notes that significant concerns have been raised about the effectiveness and availability of opt-outs. We recommend using them where available.

Monitor emerging licensing opportunities. According to the Copyright Office, individual and collective licensing agreements for AI training are emerging in certain sectors, although their availability so far is inconsistent. If your work is distributed through a label or publisher, ask what your agreement says about AI training and whether they are part of any collective licensing arrangement.

Several cases are still pending. This is their status as of November 10, 2026. We recommend checking the linked court dockets for new rulings before you rely on any decision described here.

If you believe your work is being used in AI training, we recommend discussing your specific options with an attorney who handles both copyright law and AI litigation.

For a broader overview of how copyright law applies to AI-generated content, including what you own when you create with AI tools, see our AI copyright guide for creators.


This article is for general information only — not legal advice.

Common Questions

Can AI companies legally use my music without my permission to train their AI?

Under U.S. law, an AI company does not need the copyright owner's permission for a training use that qualifies as fair use under 17 U.S.C. § 107; courts have found some AI training uses to be fair use and rejected fair use for others. AI companies have argued that training on publicly available content is fair use and requires no permission. The U.S. Copyright Office concluded in the May 2025 pre-publication version of its Part 3 Report on Generative AI Training that some uses of copyrighted works for generative AI training will qualify as fair use and some will not. Courts have reached different results: Thomson Reuters v. Ross rejected fair use for a non-generative AI tool (D. Del. 2025, affirmed by the Third Circuit in 2026), while Bartz v. Anthropic and Kadrey v. Meta (2025) found the training uses before them to be fair use on their records. Several major music copyright cases have settled through licensing deals rather than producing a final court ruling.

Book a free discovery call
Is training AI on copyrighted material fair use?

Not automatically. The fair use doctrine, codified at 17 U.S.C. § 107, requires a four-factor analysis, and no single factor is conclusive. The Copyright Office analyzed those factors in the May 2025 pre-publication version of Part 3 of its AI report and concluded that the answer will depend on what works were used, from what source, for what purpose, and with what controls on the outputs; it said commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially through illegal access, goes beyond established fair use boundaries. In Thomson Reuters v. Ross Intelligence, a federal district court rejected fair use for a non-generative AI tool built with legal headnotes, and the Third Circuit affirmed on September 29, 2026; in Bartz v. Anthropic and Kadrey v. Meta, district courts found the training uses before them to be fair use on their records. Whether that reasoning will apply to music, visual art, and other content types is still being litigated.

Book a free discovery call
What happened with the Suno and Udio lawsuits?

Record companies, including UMG Recordings, Inc., sued both platforms in June 2024. Universal Music Group settled with Udio (announced October 29, 2025), and Warner Music Group settled with Udio (announced November 19, 2025) and with Suno (announced November 25, 2025). As of November 10, 2026, Sony Music Entertainment's claims against Udio are pending in the Southern District of New York (UMG Recordings, Inc. v. Uncharted Labs, Inc.). On July 20, 2026, Sony and affiliated record labels filed a second suit against Udio in the same court (Sony Music Entertainment v. Uncharted Labs, Inc., No. 1:26-cv-06120). As of November 10, 2026, the 2024 case brought by UMG and Sony against Suno is pending in the District of Massachusetts (UMG Recordings, Inc. v. Suno, Inc., No. 1:24-cv-11611), and on September 18, 2026, UMG, Sony and affiliated record labels filed a second suit against Suno in the same court (UMG Recordings, Inc. v. Suno, Inc., No. 1:26-cv-14275).

Book a free discovery call
What did the Thomson Reuters case decide about AI and copyright?

On February 11, 2025, a Delaware federal court ruled on summary judgment that Ross Intelligence's use of Westlaw headnotes to build a non-generative AI legal-research tool was not fair use. The text is available in the court's February 11, 2025 opinion (D. Del.). On September 29, 2026, the U.S. Court of Appeals for the Third Circuit affirmed (opinion, No. 25-2153).

Book a free discovery call
How do I know if my music or art was used to train an AI?

Some AI companies have published partial information about their training datasets, but disclosure is inconsistent and incomplete. Researchers and advocacy organizations have developed tools to detect when specific images appear in known training datasets, but music and text present different technical challenges. We recommend investigating the specific model and dataset using the developer's disclosures, court records, and other lawfully obtained evidence.

Book a free discovery call
Can I send a DMCA takedown to an AI company for training on my work?

The DMCA's takedown process applies to material that resides on a system or network controlled or operated by or for a service provider and was stored there at the direction of a user (17 U.S.C. § 512(c)), and to a service provider's links or references to infringing material (§ 512(d)). A takedown notice must be a written communication to the service provider's designated agent that includes substantially the elements listed in § 512(c)(3)(A), including identification of the material claimed to be infringing or to be the subject of infringing activity that is to be removed or disabled, and information reasonably sufficient to permit the service provider to locate it; for a link or reference under § 512(d), the notice identifies that link or reference.

Book a free discovery call
How do I opt my music or artwork out of AI training?

An option to exclude your work from AI training exists only where a platform or AI developer offers one in its own terms. The U.S. Copyright Office's May 2025 pre-publication report on generative AI training notes that significant concerns have been raised about the effectiveness and availability of opt-outs. We recommend checking each provider's current terms for such an option, using it where offered, and keeping a record of its scope and the date you used it.

Book a free discovery call
Can I sue an AI company for training on my work without permission?

Yes, if you owned the copyright or an exclusive right in your work when the infringement happened (17 U.S.C. § 501(b)). For a United States work (for example, one first published in the US), you must register or preregister it with the U.S. Copyright Office before you sue. Two narrow exceptions apply. One covers the attribution and integrity rights of visual artists (17 U.S.C. § 106A(a)). The other covers sounds or images first recorded as they are transmitted (17 U.S.C. § 411(c)). At least 48 hours before that first recording, you must serve the infringer with an advance notice that complies with 37 C.F.R. § 201.22. The notice must identify the work, the time and source of its first transmission, and your intention to secure copyright. If it is a United States work, you must also register it within three months after the first transmission. If the Office refuses registration after receiving your application, deposit and fee in proper form, you can still sue. You must serve notice and a copy of your complaint on the Register of Copyrights (17 U.S.C. § 411(a)). Statutory damages are a per-work amount the court sets within legal limits. Statutory damages and attorney's fees are not available for an infringement that began before your registration took effect (17 U.S.C. § 412). Registration takes effect when the Office receives an acceptable application, deposit and fee (17 U.S.C. § 410(d)). One timing exception applies. If the infringement began after first publication, a registration within three months of that publication still allows them (17 U.S.C. § 412(2)). Section 412 also excludes three kinds of actions, named in the registration FAQ below. Late registration does not stop you from recovering the actual damages the infringement caused. You can also recover the infringer's profits from the infringement, to the extent they are not already counted in those damages (17 U.S.C. § 504(b)). The AI company can raise a fair use defense (17 U.S.C. § 107). Courts have reached both results. Thomson Reuters v. Ross rejected fair use for a non-generative AI tool, and the Third Circuit affirmed in 2026. Bartz v. Anthropic and Kadrey v. Meta (2025) found fair use for the training on their records. Bartz did not extend that ruling to pirated library copies. We recommend registering your works before any infringement begins.

Book a free discovery call
What does the U.S. Copyright Office say about AI training data?

The Copyright Office expects that some uses of copyrighted works for generative AI training will qualify as fair use and some will not. It said so in the pre-publication version of Part 3 of its Report on Copyright and Artificial Intelligence, released in May 2025. In that report, the Office concluded that several stages in the development of generative AI involve using copyrighted works in ways that implicate the owners' exclusive rights, and that the key question is whether those acts of prima facie infringement can be excused as fair use. On licensing, it recommended, at this point in time, allowing the licensing market to continue to develop without government intervention, and stated that if market failures are shown as to specific types of works in specific contexts, targeted intervention such as extended collective licensing should be considered. As of November 9, 2026, the Office has not published a final version. The report does not resolve pending litigation.

Book a free discovery call
Does registering my copyright help if an AI company used my work?

Registration does not retroactively remedy past unlicensed training. But it matters for your legal position. For a United States work, no infringement lawsuit can be filed until preregistration or copyright registration has been made, except an action under § 106A(a); if the Copyright Office refuses registration after receiving the application, deposit and fee in proper form, the applicant can sue by serving notice and a copy of the complaint on the Register of Copyrights (17 U.S.C. § 411(a)). Statutory damages and attorney's fees are not available for infringement that began before the registration took effect, unless the work was already published when the infringement began and was registered within three months after first publication (17 U.S.C. § 412). This limit does not apply to an action under § 106A(a), an action under § 411(c), or a work preregistered before the infringement began and registered within the deadlines § 412 sets. We recommend registering the copyrightable works in your catalog before future infringement begins.

Book a free discovery call
What is the difference between AI training infringement and AI output infringement?

These are two separate legal questions. Training infringement is about whether copying your work into an AI training dataset, without permission, violates your copyright. Output infringement is about whether a specific AI-generated output reproduces or closely imitates your protected expression. You can have a training claim without an output claim, and an output claim without a training claim. The Suno and Udio suits focused primarily on training. Output claims require showing that a specific generated work is substantially similar to a specific copyrighted work you own.

Book a free discovery call
Does any law require AI companies to license creators' work?

Under the Copyright Act, reproducing a copyrighted work without the owner's authorization is infringement unless a limitation in sections 107 through 122, such as fair use, applies (17 U.S.C. §§ 106(1), 107, 501(a)). In the pre-publication version of its Part 3 report (May 9, 2025), the U.S. Copyright Office stated that government intervention would be premature at this time, that licensing markets should continue to develop, and that targeted intervention such as extended collective licensing should be considered if market failures are shown as to specific types of works in specific contexts. In the EU, since August 2, 2025, the AI Act requires providers of general-purpose AI models to put in place a policy to comply with Union law on copyright and related rights, in particular to identify and comply with, including through state-of-the-art technologies, a reservation of rights expressed under Article 4(3) of Directive (EU) 2019/790, and to make publicly available a sufficiently detailed summary of the content used for training, according to a template provided by the AI Office (Regulation (EU) 2024/1689, art. 53(1)(c) and (d); art. 113, third paragraph, point (b)). Providers of models placed on the market before August 2, 2025 must comply by August 2, 2027 (art. 111(3)).

Book a free discovery call
Aghil Ebrahimi, Esq.

About the author

Aghil Ebrahimi, Esq.

Founder of StarGuard Law. Trilingual IP and technology attorney licensed in California, Ontario, and Quebec. Former touring artist and tech founder who now represents creators, founders, and agencies at the intersection of law, technology, and culture.

Work With Me

Think this applies to your situation?

Book a free discovery call