Copyright in AI-Generated Content

Is AI generated content protected by copyright in Germany?

Purely AI generated content is generally not protected by copyright in Germany. The Copyright Act (Urheberrechtsgesetz, UrhG) protects only "personal intellectual creations" (§ 2(2) UrhG), and the author is the person who creates the work (§ 7 UrhG). Copyright requires a human being whose free and creative choices are expressed in the work. Text, images, music or code produced autonomously by an AI system in response to a short prompt lack that human creative contribution.

Where a human uses AI as a tool and makes creative choices that shape the result — through selection, arrangement, editing or detailed control — the human contribution can be protected. The line between the two is the central question for businesses that create, buy or license content made with AI.

The answer matters commercially. Unprotected output can be copied freely by competitors, cannot be licensed exclusively in a meaningful sense, and offers no basis for infringement claims. At the same time, AI output can infringe the rights of others if it reproduces protected works. Companies therefore need a strategy on both sides: securing protection where possible and avoiding infringement. The questions arise across functions — marketing, product design, software development, communications and HR — so the strategy should be set centrally and implemented through policies, contracts and training rather than left to individual teams.

The human authorship requirement

German law, consistent with the case law of the Court of Justice of the European Union (CJEU), defines originality as the author's own intellectual creation reflecting free and creative choices. Both elements — a human author and creative choices — must be present. AI systems are not persons and cannot be authors; their operators or users are authors only if they contribute creative expression themselves.

Typical scenarios illustrate the spectrum:

• Short prompt, unedited output: a user asks an image generator for "a futuristic factory at sunset". The composition, style and details are determined by the model. No copyright arises in the image.
• Iterative prompting with selection: a designer runs dozens of prompts, chooses and combines elements. Selection alone is often not enough; but a creative arrangement of several outputs into a new whole can be protected as such.
• AI-assisted human work: a writer drafts a text, uses AI to suggest phrasing and edits extensively. The human-authored parts and the overall structure are protected.
• Human work processed by AI: a photographer edits a photograph with AI tools. The underlying photograph remains protected.

Prompts themselves can be protected as literary works if they are sufficiently original — long, creative prompts may qualify, simple instructions do not. Protection of the prompt does not extend to the output. The decisive question in each case is whether the human, rather than the model, determined the expressive features of the final result.

Neighbouring rights and other protection for AI generated content

Some neighbouring rights do not require creativity. Simple photographs are protected under § 72 UrhG even without originality, but this right also presupposes a human act of taking a photograph; whether purely synthetic images qualify is doubtful and not settled by the highest courts. Sound recordings, broadcasts and databases have their own rights based on investment rather than creativity: the database maker's right (§ 87a UrhG) may protect a substantial investment in obtaining, verifying or presenting content, which can include structured AI generated content collections.

Other tools may fill gaps:

• Trade marks and designs: registered designs protect the appearance of products irrespective of authorship; the applicant must be entitled to the design, which raises questions where no human designer exists. Trade marks can protect logos and brand elements regardless of how they were created.
• Unfair competition law: imitation protection under § 4 no. 3 of the Unfair Competition Act (UWG) can apply to products with competitive individuality copied in a misleading or exploitative way.
• Trade secrets: unpublished AI outputs, such as internal models, prompts and workflows, can be protected as trade secrets if kept confidential.

Infringement risks in AI generated content

Unprotected output can still infringe. If an AI system reproduces a protected work or a recognisable part of it — song lyrics, characters, photographs, code — publishing or using that output can infringe the rightholder's reproduction and making-available rights.

Adaptations require consent unless the new work keeps sufficient distance from the original (§ 23(1) UrhG). The Regional Court of Munich I held in GEMA v OpenAI (judgment of 11 November 2025 – 42 O 14139/24) that the reproduction of memorised song lyrics in chatbot outputs infringed copyright, rejecting the argument that only the user who entered the prompt was responsible, according to the court's press release. Users who publish such outputs can also be liable.

Software is a particular risk area. AI coding assistants may suggest code from open-source projects subject to licences such as the GPL. Using that code without complying with licence terms can create infringement claims and copyleft obligations; see our insight on open source compliance.

Allocating rights by contract

Because the law may not grant copyright in AI generated content, contracts become the main tool to allocate rights and risks:

1. AI provider terms: most providers assign or disclaim rights in outputs and limit warranties; enterprise terms may offer indemnities for copyright claims, subject to conditions such as use of filters. Review these carefully, including exclusions and caps.
2. Agency and freelancer agreements: require disclosure of AI use, warranties that deliverables do not infringe third-party rights, and an obligation to transfer whatever rights exist — together with an exclusivity covenant that works even where no copyright arises.
3. Customer contracts: where you deliver AI-assisted content or software, adjust warranties on originality and non-infringement accordingly.
4. Employment contracts: clarify that employees' AI-assisted work belongs to the employer and that AI policies must be followed.

Transparency: labelling AI generated content

From 2 August 2026, Article 50 of the AI Act requires providers of generative AI systems to mark synthetic audio, image, video and text output in a machine-readable format, with a transitional period until 2 December 2026 for systems placed on the market before 2 August 2026 under the Digital Omnibus on AI. Deployers must disclose deep fakes and AI-generated text published to inform the public on matters of public interest, subject to exceptions. These duties are separate from copyright, but labelling and provenance metadata also help document how content was created. See our insight on labelling AI-generated content and deepfakes.

AI generated content in marketing, media and publishing

Marketing departments and agencies are the heaviest users. Beyond copyright, marketing content must comply with unfair competition law: images or testimonials generated by AI must not mislead consumers about products, results or endorsements, and synthetic "customer reviews" are prohibited. Where AI is used to create images of real people, personality rights and data protection law apply, and consent is required in most commercial contexts. Media companies must also consider press law standards of care when AI drafts or summarises reporting, and the AI Act's disclosure duty for AI-generated text published on matters of public interest unless it has undergone human review and editorial responsibility.

For publishers and content platforms, the absence of copyright in AI output affects business models built on exclusivity. Licensing agreements with AI developers, collective licensing through collecting societies and technical measures against scraping are developing alongside the legal framework. See our insight on copyright and media law in Germany.

Internal AI policies

Companies should adopt a policy that tells employees which AI tools may be used for which types of content, which inputs are prohibited (confidential information, personal data, third-party content without rights), how outputs must be checked, and how AI use is documented and disclosed. The policy should be part of the organisation's AI literacy measures under Article 4 of the AI Act and aligned with confidentiality and data protection rules. Without such a policy, companies risk both losing protection for their own content and infringing others' rights.

International comparison

The human authorship requirement is not unique to Germany. Courts and copyright offices in other major jurisdictions, including the United States, have also refused protection for works generated without human creative control, while recognising protection for human selection, arrangement and modification. A few jurisdictions grant limited protection to computer-generated works by statute, attributing authorship to the person who made the arrangements for their creation. For international groups, this means that the same content may be treated differently across markets, which should be reflected in global licensing and content strategies.

Documenting human contributions

For content that matters commercially — brand assets, product designs, marketing campaigns, software — companies should be able to prove the human creative contribution if copyright is ever disputed. Practical measures include saving prompts, intermediate versions and edit histories; recording which employee made which creative decisions; and noting where AI suggestions were rejected or changed. This documentation is also useful for employee-invention and authorship questions and for due diligence in transactions; see our insight on IP due diligence.

Moral rights and attribution

German copyright law grants authors moral rights, including the right to be named and to object to distortion. Where AI generated content contains no human authorship, no moral rights arise. Where human authors contributed, their moral rights in their contributions remain, and contracts should address attribution and permitted modifications. Falsely attributing AI output to a human author, or a human author's work to AI, can also raise issues under unfair competition and personality rights law — the Munich court in GEMA v OpenAI dealt with, and rejected on the facts, a separate claim concerning attribution of altered lyrics.

How TRW & Co helps with AI generated content

We advise companies on copyright in AI generated content: assessing whether outputs are protectable, structuring workflows to secure human authorship, drafting AI clauses in agency, freelancer, employment and customer contracts, and reviewing AI provider terms and indemnities. We also defend and pursue infringement claims involving AI outputs and advise on AI Act labelling obligations. See our pages on the AI Act and IP and patents, our insight on text and data mining, or contact us.