A recent case concerns a claim relating to several thousand damaged items. The company in question has documented the damage caused to each individual item in numerous ‘damage reports’, including photographs.
The question arises as to how evidence of such a large number of damaged items is to be provided in legal proceedings. As regards the burden of proof under German law, the claimant would, in principle, have to provide precise evidence of each individual instance of damage to each individual item. Conversely, however, the court would also have to examine the relevant evidence in detail in order to be able to deliver a judgement. Given the sheer volume of individual claims, this proves to be extremely time-consuming and inefficient in practice, making the use of AI an obvious solution.
The damage reports can be uploaded to the AI platform internally, i.e. without any data protection concerns. Questions regarding the uploaded documents can be put to the AI. In principle, the AI is capable of scanning all files and presenting the information in summarised form. It is noticeable, however, that not all the information contained in the documents can always be extracted. For example, each document lists the damaged items with their identification numbers (in what appears to be identical formatting). However, if the AI is asked, for instance, for the total number of items listed with identification numbers in the damage reports, the AI can only extract this information for a small number of items, even though the information is actually contained in every document. In addition to the list of damaged items with their identification numbers, each document also contains the total number of items recorded in the damage report.
The AI used here also reaches its limits when it comes to image analysis. The damage reports contain all the images that visually document the damage to the individual items. According to the AI, however, the images are only available as graphics embedded in PDF files. In such a ‘mixed’ PDF file, consisting of text and images, the AI used can only recognise text content, but not image content. At present, the AI can only analyse or categorise the damage on the basis of a separate table in which all damaged items are listed again, accompanied by an additional description of the nature of the damage.
In view of the growing importance of AI in legal practice, the German Federal Bar Association (‘BRAK’) published an AI guide in December 2024 containing information on the use of artificial intelligence [1]. In this guide, the BRAK emphasises the need for lawyers to carefully review the results generated by AI themselves in order to avoid errors and the resulting liability consequences.
Section 43, first sentence, of the Federal Lawyers’ Act (‘BRAO’) governs the duty of a German lawyer to practice their profession conscientiously. Of particular importance here is the principle of highly personal service, according to which a lawyer must carry out their work independently and, in cases of doubt, personally, in accordance with Section 613 of the German Civil Code (‘BGB’). Consequently, the use of AI systems must not replace a lawyer’s work, but only support it. It is undisputed that an independent review and final verification of the AI results by the lawyer is required.
The specific duties of care when dealing with AI become stricter as the degree of automation increases and depending on the intended purpose. For example, a higher standard of care applies when AI tools are used in connection with clients (e.g. in automated communication with clients, in auto-responders, or in the use of chatbots for client onboarding) than when supporting internal workflows.
Section 43a para. 2 of the BRAO governs the prohibition on disclosing confidential information. Consequently, confidential client information must also be kept secret when using AI tools. It may only be disclosed to providers of AI tools under the strict conditions set out in Section 43e of the BRAO.
With the increasing use of AI in legal practice, the question arises for the courts as to what extent judges may rely on an AI mechanism in their assessments.
In recent years, German courts have already tested and used various programs, primarily to provide support and alleviate the workload in mass proceedings. Examples of this include the Higher Regional Court Assistant (‘OLGA’), the Artificial Intelligence Assistant for Mass Proceedings (‘MAKI’) and Codefy.
When using such AI tools, the framework established by constitutional law acts as a particular constraint; that is to say, the protection of judicial independence enshrined in Articles 97 and 92 of the Basic Law (GG), the right to one´s lawful judge under Article 101 para. 1, second sentence, of the GG, and the right to a fair trial under Article 103 para. 1 of the GG. [2]
In view of the difficulties arising in assessing the admissibility of AI tools, it seems sensible, for reasons of legal certainty, for the legislature to incorporate specific provisions on permissible areas of application into the Code of Civil Procedure, whilst taking fundamental procedural rights into account.
It might be appropriate to apply the principles governing expert evidence (‘expert reports’). The judge must be able to understand how the AI mechanism works in order to rely on its results without having to examine the documents themselves. The greatest challenge lies in the fact that neither the functioning nor the results of the systems in question are comprehensible to users. The computer functions like a ‘black box’ that is fed with data and ultimately produces a result, without it being possible to see what happens in between. Where the process is not fully comprehensible from the outside (‘black-box AI’), but its use is nevertheless necessary, the expert’s duties are limited to the design, monitoring and comprehensible presentation of the process, as well as its prerequisites and consequences, so that the court can fulfil its task (namely the assessment and evaluation of the expert report) and carry out the assessment of the evidence. Conversely, the use of AI below this threshold is, in principle, permissible, but may then entail disclosure obligations and transparency requirements.
Furthermore, the results obtained with the support of AI systems should be checked for accuracy or at least plausibility in accordance with general quality standards. Here, too, no universally applicable standards have yet been developed. However, insofar as it is part of good scientific practice to explain how results are derived and thus make them verifiable, this should also apply to the use of AI – or, given the risk of ‘hallucinations’ in large language models, even to a greater extent.
Despite the existing challenges, AI offers considerable potential in the context of legal practice for handling tasks more quickly and effectively. Nevertheless, further developments are required, and experts are always obliged to review and critically scrutinise the results generated by AI. In light of recent developments, it is to be expected that the initial problems with AI mentioned above will be resolved in future and that risks will be further minimised as artificial intelligence continues to develop. AI will significantly facilitate both the analysis and preparation of evidence by lawyers and the taking of evidence by judges.
[2] Vanetta/Vogt, Artificial Intelligence in Civil Justice – Prospects and Challenges, Der Betrieb 2025 (Issue 35), 2148 (2149); see Maddaloni, ‘Argument Mining in Civil Proceedings: Technical Potential and Necessary Regulatory Measures for Artificial Intelligence in the Judiciary’, LTZ 2025, 309 (310).
Discover our comprehensive portfolio of consulting services –
each solution carefully tailored to meet the unique needs of your organization.
Author
