AI Music in Advertising: What Is Allowed in 2026 – and What Is Not

One prompt, two minutes, and the music for the commercial is done. AI music generators such as Suno make this possible, and many marketing teams and agencies already use them. In summer 2026, however, the rules shifted: the Munich Regional Court set clear limits for Suno in a case brought by GEMA, the transparency duties of the EU AI Act have applied since 2 August, and the major labels have signed their first licensing deals with AI providers.

What does this mean for music in advertising, social media and brand communication? This guide sorts it out – with a traffic light for typical uses and a checklist for production.

As of 29 September 2026. This article summarises court rulings, legal texts and terms of use, with a focus on Germany and the EU. It is not legal advice – for specific campaigns, a legal review is recommended.

The situation in 30 seconds

• AI music in advertising is not prohibited. But it needs a clean chain of rights and sometimes a notice.
• The risk travels with the track. If a generated track sounds too close to a well-known song, it is the brand and its commercial that end up in public, not the tool.
• Purely AI-generated music usually does not enjoy copyright protection of its own. Without a human creative contribution, no copyright-protected work arises in principle – and therefore, in principle, no new GEMA repertoire. But no exclusivity arises either.
• Voices are especially sensitive. An AI voice that sounds like a real person needs that person's consent and usually a notice.

The ruling: GEMA v. Suno

On 31 July 2026, the Munich Regional Court I (case no. 42 O 763/25) largely upheld the claims of GEMA, the German collecting society, against the AI music provider Suno. The case concerned six well-known works, including “Atemlos durch die Nacht”, “Rasputin” and “Forever Young”. The court found several infringements:

• Training: The works were reproduced for training – in the US, but the claims could still be heard by a German court. The court rejected the US “fair use” defence.
• Memorisation: The models had “learned” the works in a way that they were reproducibly contained in them. That is a reproduction. The German text and data mining exception (Section 44b UrhG) does not cover it.
• Outputs: The generated songs were substantially similar to the originals. According to the ruling, Suno is responsible for this, not the users – because the prompts were simple and open-ended: lyrics, song title, musical style.

GEMA is entitled to injunctive relief, information and damages. The ruling is not yet final.

It follows a clear line: in November 2025, the same chamber had already ruled in GEMA v. OpenAI (case no. 42 O 14139/24) that memorising song lyrics in a language model is a reproduction. According to that ruling, the text and data mining exception only covers preparing the training data, not what remains inside the model. OpenAI has appealed.

What this means for advertising

In the Munich case, Suno is liable because the prompts were plain. That does not mean brands and agencies are off the hook. Anyone who publishes a track that comes too close to a protected song distributes it – with all the consequences. Prompting deliberately with song titles, original lyrics or “in the style of …” increases the risk further. Advertising is particularly exposed: wide reach, a clear commercial purpose, easy to find.

Labelling: what the EU AI Act requires since August 2026

Since 2 August 2026, the transparency duties of Article 50 of the EU AI Act apply. For audio, two roles matter:

• Providers of AI systems such as music or voice generators must mark generated audio in a machine-readable way – effective, interoperable, robust and reliable (Art. 50(2)). Systems already on the market before 2 August 2026 have a transition period until 2 December 2026.
• Those who deploy AI content, meaning the brand or agency, must disclose deepfakes (Art. 50(4)). This covers content that resembles existing persons, places, objects or events and would falsely appear to be authentic. The disclosure must be made at the latest at the first exposure to the content.

For evidently artistic, creative, satirical or fictional works, a notice that does not hamper the display or enjoyment of the work is sufficient.

What does this mean for music and voices?

• Instrumental AI music is usually not a deepfake. Art. 50(4) mostly does not require it to be labelled as AI music in the commercial. The provider's machine-readable marking should, however, be preserved.
• AI vocals or AI voice-overs that imitate an identifiable real person can fall under the deepfake rule. A corresponding notice is then required; in addition, the personality rights of the person concerned must be respected.
• Whether a commercial counts as an artistic work is an open question. When in doubt, a clear notice is the safe choice.

Unfair competition law adds another layer: an AI voice presented as a “real customer” can be misleading – regardless of the AI Act.

The European Commission's Code of Practice

How labelling should work technically is set out in the European Commission's Code of Practice, finalised on 10 June 2026. It is voluntary but serves as the benchmark: providers should combine at least two machine-readable techniques, such as signed metadata and an imperceptible watermark. For visible labelling, it provides common EU icons for AI-generated and AI-modified content.

For production, this means: metadata is easily lost in editing, export and on platforms. A watermark in the audio signal itself survives these steps much better – more on that below.

AI voices: the right to one's own voice

A ruling of the Berlin Regional Court of 20 August 2025 shows how seriously courts take the voice. The operator of a YouTube channel, who also ran an online shop, had cloned the German dubbing voice of Bruce Willis with AI and used it in two videos – without the voice actor's consent. The court found a violation of the general right of personality.

The key point: it is enough that the voice is recognised and the audience could assume the voice actor was involved. An exact copy is not required. The court rejected the “satire” argument because the use served mainly commercial purposes. As damages, it awarded a notional licence fee of 2,000 euros per video, based on usual voice-over fees for advertising.

For advertising, the conclusion is clear: recreating the voices of well-known voice actors is off limits. A “sounds a bit like …” brief to an AI tool leads into the same trap.

The clean route: licensed voices

There are now providers that license AI voices of professional voice actors – with their consent, fair pay and clearly defined usage rights. Often the same person can also be booked for studio recordings. This allows quick variations, versions in many languages or short-notice updates without legal grey areas. 42 Sounds works with such providers and selects the voice to fit the brand. More on the opportunities and duties of synthetic voices: Voice cloning for businesses.

Suno, Udio & co.: what the terms of use allow

Besides law and case law, the providers' terms decide what is allowed with a generated track. And they are changing fast.

• Suno: Commercial use is allowed on the Pro and Premier plans for properly downloaded outputs. Suno assigns its own rights in the output to the user, but expressly points out that this does not automatically create copyright protection. After its settlement with Warner Music, Suno switched to licensed models, retired older models and capped downloads.
• Udio: After its settlement with Universal Music at the end of October 2025, Udio switched off downloads. Generated songs stay on the platform. For commercials that are delivered as files to broadcasters and platforms, Udio is effectively ruled out.
• The labels: Warner has settled with Suno and Udio, Universal with Udio. Other proceedings are ongoing.

Practical tip: before every production, save the current terms of use and your own plan as a dated PDF or screenshot. Anyone later asked where a track comes from and under which terms it was created will then have an answer.

Free of GEMA fees also means: unprotected

Many see one main advantage in AI music: no GEMA fees. In principle, that is true. Music without a human creative contribution is not a work under copyright law and cannot be registered with GEMA. When registering works, authors now have to confirm that a work was not created exclusively with AI.

The downside is less well known:

• No exclusivity: Without copyright protection of its own, the music can hardly be secured exclusively under copyright law against imitation or reuse.
• Burden of proof: GEMA assumes that music used in public belongs to its repertoire. Anyone claiming “AI music, so no GEMA fees” may have to prove it.
• No guarantee of originality: No generator reliably shows whether a track contains parts of existing works.

Human plus AI: where protection arises

Things look different when a person makes the creative decisions and uses AI as a tool, for example for sound design, variations or arrangement ideas. Then a protected work can arise, with clear authorship and transferable rights. That is exactly the difference between a generated track and a brand composition.

The traffic light: what AI music in advertising is good for

An assessment based on the current situation – neither a ban nor a free pass, but guidance for planning:

• Green: pitches, mood boards, layouts, internal videosUncritical as long as the music is not published. Ideal for making directions audible quickly.
• Yellow: short-lived social clipsJustifiable with a paid plan, saved terms of use and documentation.
• Yellow: TV, radio and online commercials with media budgetOnly with a cleared chain of rights and a similarity check. Reach and run time increase the risk.
• Red: “in the style of …” prompts and soundalikes of well-known songsThis is exactly what the Munich ruling targets. Not recommended.
• Red: AI voices that resemble a real personOnly with consent, payment and a notice – otherwise off limits.
• Red: sound logo and brand music as a long-term assetWithout copyright protection of its own, the asset can hardly be secured exclusively against imitation. A sound trademark can be registered, but only protects its use as a trademark. The unclear origin of the material remains a risk. Not recommended.

Why a sound logo should be composed and what rights and costs are involved: Sound logo development.

Checking similarity: when ears are not enough

The most important question for every generated track: does it sound like something that already exists? A first check works with trained ears and recognition services such as Shazam. Both mainly find direct copies, not a melody that merely comes very close to a hit.

When in doubt, a musicological expert opinion helps. An expert compares melody, harmony, rhythm and sound with existing works and assesses whether material has been taken over. This is nothing new in the music business: such opinions are also commissioned when a well-known artist writes a new song and it needs to be checked whether it comes too close to their own earlier hits – whose rights often belong to publisher and label.

42 Sounds has worked several times with experienced musicological experts and brings them in when needed – for AI tracks as well as for traditional productions.

Checklist: 10 questions before AI music goes into a commercial

1. Which tool, which plan – and does this plan allow commercial use?
2. Have the terms of use at the time of creation been saved?
3. Were artist names, song titles, original lyrics or “in the style of …” used in the prompt?
4. Does the track sound like a well-known song? Who checked this, and how?
5. Does the track contain vocals or speech, and does the voice sound like a real person?
6. Is there consent with clear usage rights for every voice used?
7. Is a notice under Art. 50 AI Act required, and where does it appear?
8. Does the provider's machine-readable marking survive in the final commercial?
9. How long, where and in which media will the commercial run – and does the brand need exclusivity?
10. Are prompts, versions, plan and checks documented?

Proof by watermark: knowing what plays where

Labelling and proof are becoming part of production. An audio fingerprint recognises which song is playing. But it cannot tell whether the same song is currently being played for brand A or for brand B. An inaudible watermark can: it carries an ID in the signal itself – for labelling, for proof of airplay and for knowing which version is playing where.

The demo of AD-ID, the watermarking system by 42 Sounds, shows how this works in practice:

Conclusion: AI for speed, composition for the brand

AI music has arrived in advertising, and it is here to stay. For pitches, variations and fast formats it is a strong tool – if plan, terms and checks are in order. For everything that carries a brand for years – sound logo, brand music, brand voice – an original composition with a clean chain of rights remains the safe choice. It can be set up exclusively and protected by copyright, and it is free of the question of what a model once learned from.

42 Sounds combines both: sound branding and brand music from original composition, plus AI sound solutions where they save time and the rights are clear.

Sources

• Munich Regional Court I: press release GEMA v. Suno (31 July 2026, German)
• Munich Regional Court I: press release GEMA v. OpenAI (11 November 2025, German)
• EU AI Act, Article 50: transparency obligations
• European Commission: Code of Practice on marking and labelling AI-generated content
• Paul Weiss: EU Finalises Transparency Rules for AI-Generated Content
• datenschutzticker.de: Berlin Regional Court on AI voice cloning (German)
• terms.law: Suno – rights to generated music
• Digital Music News: Suno after the Warner deal
• Billboard: Udio after the Universal settlement
• purelicence.de: Is AI-generated music free of GEMA fees? (German)