AI and copyright

Training on third-party works, opt-out, rights in AI output and liability for outputs, after GEMA v OpenAI, GEMA v Suno and Like Company.

Whether AI may be trained on protected works, who owns AI results and who is liable for infringing outputs is being decided by the courts right now. We advise rightsholders, AI providers and businesses using AI.

What we do

  • Review of training data and text and data mining exceptions
  • Machine-readable reservation (opt-out) for rightsholders
  • Licences for training data
  • Protectability of and rights in AI-generated content
  • Liability for outputs and indemnity clauses
  • Enforcement against AI providers

Legal framework

Section 44b UrhG permits text and data mining unless the rightsholder has declared a machine-readable reservation. In November 2025 the Munich Regional Court held that memorisation of song lyrics in a language model is a reproduction (GEMA v OpenAI). In July 2026 followed the judgment in GEMA v Suno on AI music; an appeal is pending.

Before the CJEU, C-250/25 (Like Company v Google) is the first case on generative AI. On 3 November 2026 Advocate General Szpunar opined that chatbot answers reproducing press publications in real time may infringe press publishers' rights.

Purely machine-generated content is generally not protected by copyright for lack of human creation; the Munich Local Court confirmed this in February 2026 for AI-generated logos. Under the AI Act, providers of GPAI models must publish a copyright policy and a summary of training data.

How we proceed

  1. Analyse the use case
  2. Check rights and exceptions
  3. Licence or opt-out
  4. Contracts and policies

Four cases from practice

Anonymised and simplified. Every case depends on its facts.

Publisher sets an opt-out

A specialist publisher wanted to prohibit training on its content. We drafted a machine-readable reservation in robots.txt, metadata and terms and documented it.

Logo from an image generator

A start-up created its logo with AI. Since no copyright arises, we recommended trademark registration and human rework.

Training licence

An AI provider wanted to license photo databases. We negotiated scope, fees, deletion duties and warranties for the chain of title.

Infringing output

A company published an AI text containing passages from a textbook. We handled the warning letter and set up a review process.

Frequently asked questions

May I use AI results commercially?

Usually yes, if the provider's terms permit it and the output does not reproduce third-party works.

How do I declare an effective opt-out?

In machine-readable form, for example via robots.txt or protocols such as TDMRep, supplemented by notices in the legal notice and terms.

Who owns AI results?

Without human creative input, no one in the copyright sense. Usage rights can still be regulated by contract.

Discuss your matter

Write to us or place an order directly. No fields are mandatory.