In a ruling dated May 28, 2026 (Case No. 26 O 869/26), the Munich I District Court ruled that Google is liable for legal violations in AI-generated summaries that appear at the top of search results under the heading “AI Overview.”
1. Summary of the Decision
1.1 Facts of the Case
Google displays a summary generated by artificial intelligence (AI) above the search results under the heading “AI Overview.” In these summaries, for certain search queries that included, among other things, the word “scam”, it was stated that the plaintiffs—a publishing house and a subsidiary of that publishing house—were “known for dubious business practices” and were “often perceived as a scam”, specifically “in connection with subscription traps”. The summaries went on to mention debt collection claims, problems with digital subscriptions, lack of responsiveness, and the claim that the companies operated under various names to make identification more difficult.
These allegations were untrue. The AI-generated summary included information that did not pertain to the plaintiffs but rather to other companies with which the plaintiffs had no connection whatsoever.
The plaintiffs filed a motion with the District Court of Munich I for a preliminary injunction. The District Court of Munich I granted the motion almost in its entirety (judgment of May 28, 2026, Case No. 26 O 869/26).
1.2 Technical Background
Contrary to what the term “artificial intelligence” suggests, generative AI is not intelligent. Generative AI, which is based on so-called large language models (LLMs), has merely analyzed a vast corpus of texts during training and “memorized” how likely it is that a specific word will be followed by a specific other word in a text. Texts are assembled based on these “learned” probabilities of specific word sequences.
An AI-generated text incorporates text fragments from various sources that are in no way related to one another in terms of content. The result is not checked for factual accuracy. AI-generated texts are therefore generally free of spelling and grammatical errors. However, the content may be factually incorrect—this is referred to as “hallucinations”.
Such a “hallucination” was present in the case decided by the District Court of Munich I.
1.3 Legal Basis
1.3.1 Substantive Legal Basis
False factual claims about a company infringe upon its corporate personality rights. Companies therefore have a right to injunctive relief against anyone who disseminates false factual claims, pursuant to Sections 1004, 823(1) of the German Civil Code (BGB) in conjunction with Article 2(1) and Article 19(3) of the German Basic Law (GG).
Although in the present case the dissemination of the false factual claims was automated by an algorithm that generated and displayed the text, Google must accept responsibility for these AI-generated texts and is therefore liable as a direct interferer for an injunction, according to the Munich I District Court.
1.3.2 Procedural Background
The decision was issued in preliminary injunction proceedings; it is therefore, in any case, only provisional in nature. The decision must be reviewed in regular main proceedings, unless Google acknowledges the decision issued in the preliminary injunction proceedings as a final settlement between the parties, in which case main proceedings are unnecessary.
Google also had one month to appeal the decision to the Higher Regional Court. It is not known whether this was done.
1.4 Decision Under German National Law
The court ruled exclusively under German national law.
1.4.1 EU AI Regulation Not Applicable to Civil Claims
With regard to Regulation (EU) 2024/1689 (AI Act), the court ruled that it did not apply to this case because, to the extent that it is already in effect, it only provides for the possibility of filing a complaint with the competent market surveillance authority in addition to—and “without prejudice to”—the legal remedies already provided for under Union law and the national laws of the Member States.
1.4.2 The Digital Services Act Does Not Supersede National Laws
With regard to Regulation (EU) 2022/2065 (Digital Services Act, “DSA”), the court stated that the provisions in Art. 6, 16 et seq. of the DSA do not supersede national legal provisions either, because these provisions merely concern liability exemptions within the scope of the DSA, and furthermore, Article 6(4) of the DSA does not preclude the possibility that a judicial authority under the legal system of a Member State may require the service provider to cease or prevent an infringement.
1.5 Google Is Liable for the AI-Generated Texts as a Direct Infringer
The central issue in the proceedings was whether Google must be held responsible for the texts automatically generated by the AI.
Google had argued in this regard that the AI-generated texts did not constitute “statements” by Google, and that a claim under Sections 1004, 823 et seq. of the German Civil Code (BGB) could not be considered because Google could, at most, be held liable under the principles of indirect liability for interference, and the requirements for such liability had not been met. Google, the company argued, merely operates a search engine that automatically displays third-party data and information in response to search queries. Google is therefore not itself responsible for the data processing and does not adopt the third-party information as its own, even in the “AI Overview.” Accordingly, Google is liable only after becoming aware of a manifest legal violation.
These arguments did not convince the court. It ruled that Google is liable for the infringements as a direct infringer. In its reasoning, the court explained that the “AI Overview” is not merely a display of search results, but rather content created by Google itself and attributable to it. The fact that the term “scam” was suggested via the so-called auto-complete function as soon as one of the plaintiffs’ company names was entered likely contributed to this conclusion. Ultimately, however, the following factors were decisive:
- In response to search queries, the results are not merely displayed as links or short previews (snippets); rather, the results of the search query are summarized and evaluated in Google’s own words and according to its own structure. This presentation begins with an introductory affirmation of the query, for example with the wording “Yes, there are indications of scams and unscrupulous practices […],” and, linguistically speaking, already goes beyond the mere presentation of links.
- The independent thematic structuring of the response—into an introductory summary, followed by a compilation of the “characteristics of the suspected scam”, and then a recommendation for action (“What you can do”)—was not present at all in the cited results from third-party sites. This demonstrates that the AI independently processes the search results in terms of content. In this way, Google creates independent statements that go beyond the individual search results, which are then displayed via links later on.
- Since Google introduced the AI itself and offers it to users, it must also take responsibility for its results. After all, only Google itself has influence over the AI’s output and the algorithms it uses.
- Above all, the “AI Overview” contains statements that do not appear in the search results at all. For example, the top result in the search results was an article from a law firm titled “Beware of …”, which made no reference whatsoever to the plaintiffs. Therefore, this constitutes a statement of its own, for which Google, as the provider, must be held accountable.
1.6 Case Law of the Federal Court of Justice on the Liability and Duty of Care of Search Engine Operators for Violations of Personal Rights
The Federal Court of Justice (BGH) had already ruled in 2018 and 2013 on the search engine operator’s duty to review and liability for violations of personal rights.
1.6.1 Links in Search Results to Third-Party Websites
In 2018, the BGH ruled that a search engine operator—again, the case involved Google—is only a direct contributor to the infringement if the search results pages consist of its own content or third-party content that it adopts as its own by outwardly and recognizably assuming responsibility for the content of published material; the mere inclusion of a link in the search results is not sufficient to constitute such adoption (BGH, judgment of February 27, 2018, Case No. VI ZR 489/16). The operator of a search engine generally has no obligation to verify the legality of the content it finds. Only when the search engine operator becomes aware, through a specific report, of an obvious infringement that is clearly recognizable at first glance—such as child pornography—does it incur duties of conduct, and a breach of these duties then triggers liability as an indirect infringer.
According to the Munich I District Court, this Federal Court of Justice (BGH) precedent—which pertains to search results—does not apply to the “AI Overview” because it does not merely link to third-party websites but rather presents independent, new statements of fact based on an analysis and synthesis of content from various third-party websites. Unlike the review of content linked in search results—which would be practically impossible to carry out and would seriously call into question the existence of search engines as a business model—the Munich I District Court holds that a review of the AI’s statements is entirely possible, at least by comparing the underlying third-party websites with the AI’s own statements based on them.
1.6.2 “Autocomplete” Function
Similarly, the District Court of Munich I also rejected the application of the Federal Court of Justice’s case law regarding Google’s so-called “autocomplete” function. In 2013, the Federal Court of Justice had ruled that Google is liable for the display of terms infringing on personality rights in the “autocomplete” function when the name of the affected person is entered only if it fails to fulfill reasonable duties of scrutiny, and that such liability arises only upon becoming aware of the infringement of personality rights (BGH, judgment of May 14, 2013 Case No. VI ZR 269/12).
According to the Munich I District Court, the “AI Overview” goes far beyond the display of term associations formed from users’ search data, because it involves an independently generated statement that, based on the weighting of search results and an evaluation of the content, formulates its own answer (not a question) and presents it to the user.
1.7 Users’ Ability to Verify Is Irrelevant
In the opinion of the Munich I District Court, the fact that users can use the links to verify whether the content of the third-party sites actually corresponds to the “AI Overview” does not relieve the provider of liability for the statement, because the “AI Overview” is comprehensible in and of itself, constitutes a complete statement with independently understandable content, and contains no indication of alternative interpretations or, indeed, any unreliability in its content. Therefore, users generally have no reason to additionally verify the displayed answer to the search query.
1.7 Users’ Ability to Verify Is Irrelevant
In the opinion of the Munich I District Court, the fact that users can use the links to verify whether the content of the third-party sites actually corresponds to the “AI Overview” does not relieve the provider of liability for the statement, because the “AI Overview” is comprehensible in and of itself, constitutes a complete statement with independently understandable content, and contains no indication of alternative interpretations or, indeed, any unreliability in its content. Therefore, users generally have no reason to additionally verify the displayed answer to the search query.
1.8 Functionality of the Search Engine Not Impaired
While the Federal Court of Justice had cited the functionality of the search engine as grounds for denying a duty to verify without cause in the case of links in search results, the District Court of Munich I also rejected this line of reasoning because the “AI Overview” merely offers an additional feature, without which the search engine can still be used, and without which users are certainly still able to find results in the “flood of data” .
2. Opinion
The result seems surprising: The liability exemption in the form of a notice-and-take-down procedure, which German case law has thus far recognized in cases of third-party infringements not only in favor of search engine operators but also in favor of hosting providers (BGH, judgment of October 25, 2011, Case No. VI ZR 93/10), is not supposed to apply if the “third party” is an AI.
However, the AI is deployed by Google, and its results are made available to Google users. The AI is therefore not a third party but falls within Google’s sphere of responsibility. As the court correctly notes, the search engine would function and could be used even without the “AI Overview.” After all, Google has operated the search engine extremely successfully for more than two decades without the “AI Overview.”
This raises the question of how Google will respond to the decision. Completely abandoning the “AI Overview” will certainly not be the solution. Google invests billions in AI development. As the ruling indicates, after receiving the cease-and-desist letter, entering the same search terms displayed different summaries that no longer contained the disputed statements. Apparently, Google was able to take measures to ensure that the disputed statements were no longer displayed.
The court’s rejection of the “notice-and-take-down” procedure raises the question of how Google should prevent inaccurate statements made by AI in the future without prior warning or other notification of the legal violation. One (largely) legally sound option would be to program the AI to refrain from making any negative statements about companies and individuals. However, this would result in users being deprived of important information—for example, if they receive a suspicious email and ask Google whether it is a phishing email. An AI that generally omits any mention of problems of any kind in order to avoid legal violations is completely useless. This is because people typically consult an AI precisely in situations where they are facing a problem or need to make a decision and require information to resolve it—information that they would be unable to find, or would find only with great difficulty, without the AI.
In this context, the court’s reasoning is interesting: it held that the “AI Overview” contains no indication of alternative interpretations or even inaccuracies in the content, and that users therefore generally have no reason to further verify the displayed response to their search query. An initial measure that could be implemented quickly would therefore be to include a prominent disclaimer at the beginning of the AI-generated summary.
Currently, the “AI Overview” includes a note at the end in gray text that is smaller than the main text, which reads as follows: “AI responses may include mistakes. Learn more” The text “Learn more” is linked to a page with information about the “AI Overview.” For search queries on legal topics, the notice additionally includes the sentence “For legal advice, consult a professional.”. For search queries on medical topics, the sentences “This is for informational purposes only. For medical advice or diagnosis, consult a professional.” are added.
It is not clear from the judgment of the Munich I District Court whether these notices were already in place at the time of the infringement. In any case, these notices are a step in the right direction; however, it is doubtful whether they are sufficient—particularly given their placement at the end of the overview and their inconspicuous design—to trigger the notice-and-take-down procedure.
AI developers will have to devote greater attention to the issue of preventing legal infringements. This applies in particular to cases of “imposed AI”, in which the user does not consciously intend to use an AI (such as Gemini, ChatGPT, or Copilot) but rather a traditional search engine that has existed for over 25 years, and is therefore potentially less aware of the risk of misinformation than an experienced AI user who consciously uses an AI.
Content generated using artificial intelligence may enjoy copyright protection under certain circumstances. This was decided by the Munich Local Court in February (final judgment of February 13, 2026, file no. 142 C 9786/25). In the specific case decided, however, the court denied copyright protection. In recent years, US courts had already ruled that AI-generated content is not eligible for copyright protection under US law (e.g., U.S. Court of Appeals D. Columbia Circuit, judgment of March 18, 2025, file no. 23-5233 – Stephen Thaler v. U.S. Copyright Office).
Facts
The plaintiff had created the “logos” reproduced below using generative AI and used them on his website.
The defendant reproduced these logos and used them on his website without the plaintiff’s consent.
Argumentation of the Plaintiff
The plaintiff argued that the AI used to create the logos was merely a tool – albeit a particularly powerful one – whose use did not preclude personal intellectual creation within the meaning of Section 2 (2) of the German Copyright Act (UrhG) if the design of the product could be attributed to the intellectual creative act of a human being. An essential indication of the act of intellectual creation was the iterative process and the multiple human editing of the product, through which the product was increasingly shaped in accordance with direct human input. The iterative revision process can be compared to the work of a sculptor who carves a statue out of stone step by step and checks at each intermediate stage whether the current state of his work already corresponds to his intellectual creative conception and, if necessary, makes corrective interventions.
Argumentation of the Defendant
The defendant argued that the logos could not constitute works eligible for copyright protection within the meaning of Section 2 of the German Copyright Act (UrhG) because they had not been created by a human being. The user of generative AI does not perform any creative work. The very fact that anyone can produce complex, stylistically diverse works en masse in a very short time and with minimal effort shows that the creative contribution of humans is completely overshadowed by automated generation. The lack of control by the user over the creative process and the lack of predictability of the specific output also argue against the mere tool characteristic. The AI decides for itself what the product should look like, without the user having any control over this.
Court Decision
The Munich Local Court dismissed the action for an injunction against the publication of the logos and for their removal from the defendant’s website. It ruled that the products in dispute were not works of applied art protected by copyright under Section 2 (1) No. 4, (2) UrhG.
Copyright Protection is possible with Human Creative Influence
The Munich Local Court did not rule that products generated by artificial intelligence generally cannot have the character of a work within the meaning of copyright law.
Rather, the court focused on the extent to which human creative influence is still exerted despite the software-controlled process flow. Copyright protection is therefore conceivable as a result of human intervention in AI results, whereby this intervention can also take place retrospectively or successively during prompting if it leads to the output also reflecting the personality of the person prompting.
Human creative influence on the design of the specific work itself is therefore necessary, for example through sufficiently individual presettings when programming the creation process of the specific product itself, possibly in conjunction with a selection process among the generated products. The mere selection of an AI product from several “suggestions” made by the AI is not sufficient in itself. Copyright protection and ancillary copyright protection for the AI product are not applicable if the generation of the product is entirely software-controlled.
Creative Elements in Prompting must Dominate the Output
Ultimately, the decisive factor is whether the prompting expresses the creative abilities of the person prompting in an independent manner by making free and creative decisions and thus also giving the output a personal touch. The design should not be dictated by the technical function of the AI, but rather the person prompting must express their creative spirit in an original way.
Figuratively speaking, it is necessary that the use of the AI model is closer to a tool than an independent creative instrument. Ultimately, the input must shape the resulting output in a sufficiently objective and clearly identifiable manner.
In the court’s opinion, this is only the case if the creative elements incorporated into the prompt dominate the output to such an extent that the object as a whole can be regarded as the author’s own original creation.
It is therefore not sufficient if, in the context of prompting, the creative “decision” is ultimately left to the AI by means of only general, open-ended instructions, even if these are numerous and thus gradually change the appearance of the output.
Costs and Effort are Irrelevant
The Munich Local Court rejected the plaintiff’s argument that the costs, effort, and care involved in creating the prompt should be taken into account in the assessment. According to the court, it is completely irrelevant whether a paid premium version of the AI is used or how complex and carefully a prompt was created. This is because purely manual activities do not reflect personality, regardless of how costly or time-consuming these activities are. Copyright law does not reward and protect investment, effort, or diligence, but only the result of a creative activity.
Analysis of the Prompts
The plaintiff had documented the individual prompting steps he had used to create the logos and presented them in court. The court analyzed this documentation in detail and concluded that none of the three logos could be considered an original work of the plaintiff in which his personality was expressed as the result of a free creative decision.
Presettings in the Programming of the Creation Process
It is not entirely clear what the court means by “presettings when programming the creation process of the specific product itself”, which could justify human creative influence on the design of the specific work. If this means that the author was already involved in programming the AI system—as opposed to merely using an existing AI system—then this case is likely to be very rare in practice.
Protection of Investment and Effort
The Munich Local Court’s remark that copyright does not reward and protect the investment, effort, or diligence, but only the result of creative activity, applies only to copyright, not to the related rights regulated in Sections 70 et seq. of the German Copyright Act (UrhG). In the context of related rights, such as the protection of database producers pursuant to Sections 87a et seq. UrhG, investment and effort are indeed protected.
Significance of the Decision
The decision is of fundamental importance in that the court did not a priori reject copyright protection for products created using AI. As is so often the case, the same applies here: it depends. Even a work created with the help of AI can enjoy copyright protection if it is the result of a creative process by a human creator and the latter has clearly and objectively identifiably influenced the output.
According to these principles, for example, an image generated by AI based on the alteration and processing of a photo, to which the AI user holds the copyright, and which was previously uploaded to the AI by the AI user, should be eligible for copyright protection without further ado.
The decision is also interesting in light of the Ad Blocker IV (Werbeblocker IV) decision of the Federal Court of Justice (BGH) of July 31, 2025 (file no. I ZR 131/23), see Ad‑blockers infringing, cheat software not – TCI Rechtsanwälte. In that case, the Federal Court of Justice included code in the scope of copyright protection for computer programs that was not created directly by the author of the computer program, but only indirectly, in that the author created code that in turn generates code. The Federal Court of Justice thus recognized that copyright protection cannot be denied solely on the grounds that the work in its concrete form is the output of a computer.
However, protection for a logo created using AI can be obtained by registering the logo as a trademark. You can read about what needs to be considered in detail here: AI-generated content and trademark law – TCI Rechtsanwälte.
The 2nd edition of the Commentary on IT Law, to which our partners Dr. Truiken Heydn, Dr. Michael Karger, and Dr. Thomas Stögmüller contributed as authors, will soon be published by Otto Schmidt Verlag. The commentary can be accessed online via juris.
On October 30, 2025, our partner Dr. Truiken Heydn will speak at the European Conference of the International Technology Law Association (ITechLaw) in London, together with Eugene Weitz and John Beardwood, on important contract clauses that may not be on everyone’s radar.
This will not be about the clauses on warranty and liability. If you would like to know more, you can still register here, but the number of participants is limited.
As of today (September 12, 2025), Regulation (EU) 2023/2854 (Data Act) applies.
What does the Data Act govern?
The Data Act governs data rights. This includes, among other things, data generated by connected devices in the context of Industry 4.0 applications and the Internet of Things (IoT).
Connected devices placed on the EU market must be designed in such a way that they enable the sharing of data. In addition, consumers must have the option of choosing less expensive repair and maintenance service providers or performing these tasks themselves, i.e., they must not be forced to commission the manufacturer to carry out repairs and maintenance by being denied access to data required for repair and maintenance.
Commercial users of equipment must have access to the data generated by that equipment in order to improve its efficiency and operation.
In particular: Cloud Computing
The Data Act is of particular importance for all providers and users of cloud computing. Cloud users must be able to switch to another cloud provider and migrate their data from their previous cloud provider to another cloud provider.
In particular, cloud providers are prohibited from making the switch dependent on payments, thereby effectively preventing the switch. However, until January 12, 2027, providers of data processing services may still charge reduced switching fees, but not after that date.
Model contractual terms and standard contractual clauses missing
Article 41 of the Data Act states that the Commission shall “before September 12, 2025” develop and recommend non-binding model contractual terms for data access and data use, as well as non-binding standard contractual clauses for cloud computing contracts.
However, these are not yet available. In the Commission’s press release of September 12, 2025, it states under “Next Steps” that the Commission will publish model terms for data sharing and standard clauses for cloud contracts – but it is unclear when.
What is available so far is the final report of the Commission’s expert group dated April 2, 2025, which contains proposals for model contractual terms and standard contractual clauses. However, this is not available in German, and it is also unclear whether the clauses proposed therein will ultimately remain unchanged.
A novelty outside Germany: B2B general terms and conditions law
Finally, special attention should be paid to the fact that the Data Regulation introduces, for the first time at EU level, a law on general terms and conditions that also applies to relationships between businesses (B2B). According to Article 13 of the Data Act, contractual terms concerning access to and the use of data or liability and remedies for the breach or the termination of data-related obligations, which have been unilaterally imposed by an enterprise on another enterprise, shall not be binding on the latter enterprise if they are unfair.
Until now, EU-level law on standard terms and conditions has only applied to relationships between businesses and consumers (B2C), such as Directive 93/13/EEC on unfair terms in consumer contracts.
However, the law on standard terms and conditions applicable in Germany is also largely applied by the Federal Court of Justice (BGH) in relations between companies, which is met with great incomprehension internationally in most cases. In Germany, EU directives such as Directive (EU) 2019/770 on certain aspects concerning contracts for the supply of digital content and digital services, have been implemented in Germany in an “excessive” manner in that the regress provisions in the supply chain upstream of consumer transactions in Section 327u of the German Civil Code (BGB) have been made mandatory between companies, whereas according to the directive, only the provisions relating to consumers need to be mandatory.
On August 30, 2025 our partner Dr. Truiken Heydn gave an interview on Deutschlandfunk Kultur radio in the program “Breitband” about the German Federal Court of Justice ruling of July 31, 2025, on the copyright infringement of ad blockers. You find the interview here (sorry, this one is in German language):
New BGH Case Law on the Modification of Computer Programs
In two rulings delivered on July 31, 2025, the First Civil Senate of the German Federal Court of Justice (Bundesgerichtshof – BGH), responsible for copyright law, addressed the issue of modifying computer programs.
Ad‑blocker (Judgment of July 31, 2025, File No. I ZR 131/23 – Werbeblocker IV)
An operator of several online portals contested against the distributor of a browser plug‑in designed to suppress advertisements on websites.
How the ad‑blocker works
The ad‑blocker interferes with data structures generated by the browser in a user’s RAM after fetching and loading the HTML file. It alters the object structure (the DOM node tree) as well as the formatting structures (CSSOM) used by the website, preventing elements recognized as advertisements from appearing on the user’s screen. This happens either by preventing ad content from being retrieved from ad servers or by loading it into memory without displaying it.
Copyright infringement through interference with protected code
The BGH held that the exclusive right to modify a computer program under § 69c No. 2 first sentence of the German Copyright Act (UrhG) and the exclusive reproduction right under § 69c No. 1 first sentence UrhG may be infringed if the browser and its engines are not controlled via object code, but by bytecode from which the browser’s virtual machines generate object code. That bytecode—or the code generated from it—may be protected as a computer program, and an ad‑blocker that alters this code in the course of reproduction may infringe the exclusive right associated with it.
Decision
The BGH referred the case back to the appellate court for a new hearing and decision, so that it can clarify which part of the website (bytecode or object code) is affected, whether it is protected by copyright, and under what conditions the interference may be justified.
Cheat Software (Judgment of July 31, 2025, File No. I ZR 157/21 – Action Replay II)
The distributor of PlayStation consoles and associated games challenged the “Action Replay PSP” software, which enables users to manipulate original games to bypass restrictions such as the limitation of the use of “boosters.”
How the cheat software works
Using “Action Replay PSP” requires connecting the PSP to a PC and inserting a Memory Stick. After rebooting the PSP, users can access an additional menu labeled “Action Replay” to deactivate in‑game limitations. The software runs concurrently with the game software and does not modify the object or source code or the internal structure of the game. Instead, during gameplay, it alters the values of variables that the game software stored in the PSP’s RAM and uses during execution. As a result, the game runs based on these modified variable values.
No copyright infringement without interference with protected code
The BGH held that the exclusive rights under § 69c No. 2 and § 69c No. 1 UrhG are not infringed when the cheat software does not alter the program data of the object or source code of the game, but only other elements such as variable data created in RAM during program execution. If only the program flow is influenced in this way, it does not constitute an infringement of the software copyright.
Significance of the Decisions
On the one hand, an ad‑blocker may infringe website copyright, even though it merely prevents certain content from being displayed. On the other hand, software that modifies the intended program flow of a video game is classified as non‑infringing. At first glance, this may appear surprising.
Criterion: Interference with code
The BGH draws a legal distinction based on whether the code is interfered with. The author of a computer program creates the code, and only if this code is modified is their right violated. In the Action Replay II case, the BGH found no modification to code, while in the Werbeblocker IV case, it deemed that not only the author-created source and object code may be protected, but also a bytecode running on a virtual machine that translates it into machine code.
As an aside: Programming languages without object code and platform‑independent programs
For instance, Java programs are not directly compiled into object code. Instead, they are compiled into bytecode executed on a virtual machine (part of the Java runtime environment), which translates the bytecode just in time into machine code for execution. This enables platform independence, provided the appropriate runtime environment is installed.
Extension to code not directly created by the author
Interestingly, the BGH includes code under copyright protection that was not directly written by the program author, but indirectly generated from code created by the author.
Application to AI‑created computer programs?
Application to AI‑created computer programs?
This development may be relevant for future considerations. It suggests that programs created using artificial intelligence might also be protected by copyright. This raises further questions: Who should be regarded as the author—the one who programmed the AI model, programmed the AI system, trained the AI system, or created the training data?
Our partner Dr. Truiken Heydn was elected to the Executive Committee of the International Technology Law Association (ITechLaw) as of May 2025. She will initially fill the position of Assistant Secretary there. She previously served on the Board of Directors from 2013 – 2019. ITechLaw is the leading international association for lawyers specializing in technology law.
On February 12, 2025, the EU Commission withdrew the AI Liability Directive. The Commission justified its decision by stating that no foreseeable agreement was reached on the directive.
Proposed AI Liability Directive
The proposed directive on adapting non-contractual civil liability rules to artificial intelligence (AI Liability Directive) was intended to establish certain rules for the assertion of non-contractual civil claims for damages caused by artificial intelligence. According to a survey conducted in 2020, the liability risk is the most important obstacle preventing companies from using AI.
The AI Liability Directive was part of a package of measures to support the introduction of AI in Europe, alongside the AI Act and the revision of EU product safety law. The AI Act is in force; Chapters I and II (General Provisions and Prohibited AI Practices) apply since February 2, 2025. The Product Safety Regulation is also in force; it applies since December 13, 2024.
Strict liability?
The proposed AI Liability Directive did not provide for strict liability for AI systems.
However, it should be noted that the new Product Liability Directive now expressly clarifies in Art. 4 No. 1 that software is considered a product within the meaning of the Directive regardless of its embodiment on a physical medium. This means that strict product liability also applies to AI systems, as every AI system is based on software. Until now, it has been disputed whether software constitutes a product within the meaning of product liability law.
Obligation to disclose evidence
The proposed AI Liability Directive was intended to significantly reduce the burden of proof for those injured by an AI system. The provider or user of a high-risk AI system should be obliged to disclose relevant evidence. The civil courts should be given the power to order this disclosure in damages proceedings.
Reversal of the burden of proof
The refusal to disclose should then give rise to a rebuttable presumption of a non-compliance with a relevant duty of care by the provider or user.
Furthermore, even in the case of AI systems that are not high-risk AI systems, a reversal of the burden of proof for the causal link between the defendant’s fault and the damage should apply under certain conditions: In these cases, the defendant should have to prove that it is not responsible for the damage.
These regulations will not come into existence for the time being. It remains to be seen whether the plan for an AI Liability Directive will be taken up again by the Commission at a later date.
Significance of the Decision
The Commission’s decision has been welcomed by the digital industry in particular.
No disclosure of evidence
From the point of view of the administration of justice, the decision is also to be welcomed.
The obligation of the Member States to introduce a power for the courts to order the disclosure of evidence represents a significant encroachment on the civil procedural law of the Member States. This applies in particular to Member States such as Germany, under whose civil procedural law the principle of production of evidence applies, i.e. the evidence must be produced by the parties and the court does not investigate the facts of the case or obtain evidence of its own motion. The principle of production of evidence is an outgrowth of the party autonomy applicable in civil law and is therefore ultimately one of the fundamental civil liberties.
No Reversal of the Burden of Proof?
However, the Commission’s decision should not be overestimated with regard to the regulations on the reversal of the burden of proof, which will not come into existence for the time being.
This is because the German Federal Court of Justice established a reversal of the burden of proof with regard to fault as early as 1968 within the framework of the generally fault-based non-contractual producer liability pursuant to Section 823 (1) of the German Civil Code. As a result, there are already hardly any differences in practice in Germany between liability under Section 823 (1) of the Civil Code and strict liability under the Product Liability Act, which came into force in 1990.
There is also a reversal of the burden of proof for fault in contractual liability in Germany (Section 280 (1) sentence 2 of the Civil Code).
These rules on the reversal of the burden of proof naturally also apply in claims for damages in which an AI was involved in causing the damage.
Conclusion
The Commission’s decision not to harmonize this area for the time being and to leave the regulation of AI liability to the Member States is a step in the right direction. After all, given the virtual dominance of US companies in this area, it is questionable whether the sometimes excessive regulation of IT law at the EU level really helps the European digital economy.
