How Musicians Are Fighting Back Against AI Music Theft

When Suno and Udio launched their AI music generators to the public in 2024, they opened a consumer product built on a foundation that the music industry immediately identified as deeply problematic: training data that almost certainly included copyrighted recordings scraped without licensing, consent, or compensation. Two years later, the legal battles are in full swing, new technical tools are being deployed by artists to protect their work, and a broader reckoning about who benefits from AI trained on human creativity is reshaping the music industry. Here's where things stand.

The Training Data Problem

AI music generators are trained on vast datasets of audio. The quality of a generator's output depends heavily on the quality and volume of its training data — which means the best outputs come from models trained on the best professional recordings. Those recordings are copyrighted. The core legal argument by artists and labels is straightforward: using copyrighted music to train a commercial AI product without licensing or consent is copyright infringement, regardless of whether the model's outputs directly reproduce those recordings.

The counterargument from AI companies has been "transformative use" — arguing that training is analogous to a human musician listening to and learning from music, and that the AI's outputs are transformative works rather than reproductions of the training data. Courts are working through this argument now, and the outcomes will define AI music law for the foreseeable future.

The Lawsuits

The Recording Industry Association of America (RIAA) filed suit against Suno and Udio in June 2024, representing Universal Music Group, Sony Music, and Warner Music Group — a coalition covering the vast majority of commercially significant recorded music. The suits allege copyright infringement for using copyrighted recordings in training data and seek statutory damages that, if applied to the full scale of the alleged infringement, could reach into the billions of dollars.

In parallel, individual artists have filed or joined class action suits. These include prominent recording artists who allege not just copyright infringement in training but additional claims: right of publicity (their voice and style being commercialized without consent); trade dress (AI-generated music mimicking their recognizable sound); and unfair competition. The class actions aggregate the claims of many artists whose work was used in training data, increasing the potential damages and the breadth of the legal challenge.

Outside the US, the EU's AI Act creates explicit requirements around transparency about training data and establishes rights for rights holders to opt out of having their work used in AI training. Several European artists and collecting societies have brought actions under this framework.

What Artists Are Doing Practically

Beyond the courts, artists are pursuing several practical strategies.

Opting out of training datasets. Several AI companies have implemented opt-out mechanisms that allow rights holders to request their work not be used in future training. The practical limitations are significant: opt-outs are prospective, not retrospective (they don't affect models already trained on the work); they require the rights holder to identify their work specifically; and compliance is voluntary and unverifiable. Many artists consider opt-outs inadequate given that the initial training happened without consent.

Using "poisoning" tools. Researchers at the University of Chicago developed a tool called Nightshade that subtly modifies images in ways invisible to the human eye but that corrupt AI training — causing models trained on poisoned images to produce degraded, incorrect outputs. Similar tools are being developed for audio. AudioShield and similar projects aim to modify the spectral properties of released recordings in ways that don't affect perceptual quality but interfere with how AI models can learn from them. The practical efficacy of these tools at scale remains debated, but they represent a novel technical countermeasure.

Demanding transparency. Several artist advocacy organizations — including the Artist Rights Alliance and the Future of Music Coalition — are pushing for mandatory disclosure of training data. Under their proposals, any AI music generator sold commercially would be required to maintain and publish a detailed record of all music used in training, enabling rights holders to verify whether their work was included and pursue appropriate remedies.

Collective licensing negotiations. Some collecting societies and artist advocacy groups are negotiating industry-wide licensing frameworks rather than pursuing individual claims. The logic is pragmatic: even if AI companies are legally required to license training data, individual artists lack the resources and leverage to negotiate meaningful deals. Collective arrangements — similar to how performance rights organizations (PROs) like ASCAP and BMI operate — could provide broader coverage and revenue distribution.

The Streaming Platform Angle

Streaming platforms have a complicated relationship with AI music. On one hand, AI-generated music flooding streaming catalogs dilutes the per-stream royalties paid to human artists (since total royalty pools are divided across all streams). Spotify, Apple Music, and others have implemented policies requiring disclosure of AI-generated content and, in some cases, restricting AI-generated music from certain high-visibility placements. On the other hand, platforms benefit from increased catalog volume.

The royalty dilution problem is particularly acute. An AI generator that can produce unlimited content, distributed across millions of tracks under fake artist names, can theoretically capture a significant portion of streaming royalties at near-zero marginal cost. Several streaming platforms have identified and removed coordinated AI music manipulation operations — but the detection and removal is reactive and the scale is difficult to manage.

AI and Artist Identity

A separate and arguably more personal concern for many artists is AI voice cloning and style mimicry — using AI to generate music "in the style of" or directly mimicking the voice of specific artists. When an AI tool can generate a song that sounds like it could have been made by a specific artist, that artist's brand, reputation, and commercial value are affected — even if no specific recording was directly reproduced.

Several high-profile cases have brought this into public attention: AI-generated "new tracks" from deceased artists, voice-cloned covers that circulated as genuine, and AI-generated tracks released under fake artist names that mimicked real artists' identities closely enough to confuse fans. Right of publicity law provides some protection against commercial use of an artist's voice and likeness without consent, but enforcement against anonymous AI-generated content at internet scale is practically difficult.

The Consent and Compensation Framework

The framework that artist advocates are pushing toward has three pillars: consent (artists must affirmatively agree for their music to be used in AI training), transparency (training data must be documented and disclosed), and compensation (artists must receive fair payment for training data use). This framework is already partially reflected in the EU AI Act and is the basis for active legislative proposals in the US, UK, Canada, and Australia.

Whether this framework will be established through litigation, legislation, or negotiated industry standards — and on what timeline — is unclear. The court cases will not resolve quickly; major copyright litigation typically takes years. Legislation in the US has moved slowly. In the interim, the AI music industry continues to operate, artists continue to push back through every available channel, and the commercial stakes grow with every new model iteration.

What Listeners Can Do

Music consumers have more agency in this debate than they might think. Choosing to listen to and support human artists — by streaming their verified music, attending live performances, and purchasing directly from artists — keeps royalties flowing to human creators. Learning to identify AI-generated music (using tools like SUS IT, or developing the ear for the tells described in our detection guide) and avoiding AI music packaged deceptively as human-made is a form of active participation in the debate about what we value in music. The music industry's future will be shaped partly by what audiences choose to reward.