Handpicked decisions with immediate relevance to our mandates — curated and annotated, with one entry added every Friday. Here you will find the ten most recent decisions in full length; older entries are linked in the archive.
// As of: May 2026 · 11 entries · the 10 most recent in full length
Where a language model outputs memorized song lyrics almost verbatim, the operator is liable — not the user.
The court treated the output of memorized texts as reproduction (§ 16 UrhG) and making available to the public (§ 19a UrhG); the TDM exception (§ 44b UrhG) does not apply because the content is stored permanently. Responsibility lies with the model operator. An appeal has been announced; not final and binding. Practical takeaway: Anyone deploying generative AI should contractually clarify licensing and liability questions regarding training and output content with the provider — responsibility cannot simply be shifted onto users across the board.
View source →
AI training datasets are permissible under the TDM exceptions — an opt-out formulated only in natural language was not "machine-readable" in 2021.
The OLG confirmed the permissibility of dataset creation (§§ 44b, 60d UrhG) but denied that the reservation of use was machine-readable for 2021, leaving open whether natural language would suffice today. An appeal on points of law (Revision) to the BGH has been admitted and filed. Practical takeaway: Rightholders should set opt-out reservations in technically machine-readable form; for AI providers, the legal position on training remains in flux until the BGH decision.
View source →
Where the employer permits voluntary ChatGPT use via private accounts, there is no co-determination right — its introduction as a workplace tool, by contrast, does trigger one.
As the employer had no access to the data accruing at the provider, the court denied a co-determination right under § 87(1) no. 6 BetrVG; the employer's introduction of AI as a technical device, by contrast, is subject to co-determination. Practical takeaway: Anyone rolling out AI tools across the business should involve the works council and govern usage in an AI policy or works agreement.
View source →
An AI cannot be named as an inventor — but AI-assisted inventions are patentable where a natural person is named as the inventor.
An inventor within the meaning of § 37 PatG can only be a human being; the use of an AI to find the technical teaching does not preclude patentability, as long as there is a substantial human contribution. Practical takeaway: In AI-assisted research and development, document the human inventive contribution and name it correctly in the application.
View source →
Expulsion clauses against manager-shareholders are valid where the shareholding is tied solely to the corporate office.
The BGH adheres to the basic rule that expulsion without compensation is contrary to public policy (§ 138 BGB), but permits call options on pure management shareholdings that carry no independent weight. Pricing and compensation questions do not affect the validity of the clause but are assessed separately. Practical takeaway: Leaver and call-option clauses in PE/VC and management-participation agreements can be drafted with legal certainty where the shareholding is cleanly coupled to the corporate office.
View source →
Synchronous live online coaching does not fall under the Distance Learning Protection Act (FernUSG).
What is decisive is the content of the contract, not how it is actually carried out. Live calls with a genuine opportunity for interaction remain valid — programs that are predominantly pre-recorded, by contrast, are void. Practical takeaway: A reclaim is worthwhile only after a precise analysis of the format; a blanket "contract void" no longer holds.
View source →
The FernUSG also protects businesses — coaching contracts without ZFU authorization are void.
Structured online mentoring programs with monitoring of learning success require authorization. Where authorization is lacking, the contract is void from the outset under § 7 FernUSG and amounts paid are fully recoverable. Practical takeaway: B2B customers, too, can invoke this protective statute.
View source →
Automated scoring is a "decision in the individual case" under Art. 22 GDPR.
The mere calculation of a score value already falls under the prohibition on automated individual decisions where third parties decisively rely on it. Practical takeaway: Anyone deploying AI-assisted assessment systems needs a legal basis, transparency about the logic involved, and a right of intervention for the data subject — a mandatory topic for every AI implementation.
View source →
Even the mere loss of control over data gives rise to non-material damages.
Concrete misuse of the data is not required; the BGH considers around EUR 100 per affected person to be reasonable. Practical takeaway: Data protection incidents scale into substantial risks through mass claims — sound data management and incident processes become a liability question.
View source →
A shareholder resolution that breaches the articles of association is merely voidable, not void.
Where a resolution breaches the allocation of competences in the articles of association — rather than mandatory law — it remains valid until successfully challenged. Practical takeaway: In GmbH disputes and when removing managing directors, the one-month period is decisive; whoever lets it lapse loses the resolution.
View source →