Court cancels fee: AI use leads to expert opinion being unusable

In the legal world, the commissioning of expert witnesses was previously generally considered a guarantee of professional depth and human expertise. However, the increasing integration of Artificial Intelligence (AI) into the judiciary is now also leading to undesirable consequences in the courtroom. The Darmstadt Regional Court, with a decision published on November 10th, has set a signal against the non-transparent use of AI systems in judicial expert opinions (Ref.: 19 O 527/16). In this case, it reduced the remuneration of a specialist in oral and maxillofacial surgery to exactly zero euros.

According to the decision, the expert witness had presented the court with an invoice for 2374.50 euros. However, the responsible civil chamber refused payment entirely, classifying the submitted, extremely concise work as legally unusable.

The court's reasoning reads like an analysis of the specifics of current large language models like GPT, Claude, or Gemini. The judges are convinced that the "expert opinion," which they only placed in quotation marks, was generated in essential parts using AI. However, the commissioned professor had not disclosed this to them.

The Machine Leaves Traces

Several factors typical of AI-generated texts were decisive. The lawyers noticed bizarre formulations in which the expert witness named himself, including his full address, as the recipient of the order for evidence. The monotonous structure of the text, which "consisted almost exclusively of main clauses" with identical sentence beginnings, as well as atypical repetitions of file numbers and dates, were also interpreted by the judges as indicators of a machine pattern.

What is particularly striking about the case is the expert witness's obvious negligence in reviewing the AI results. The court refers to fragments in the text that can best be explained by insufficient correction of the prompts. For example, a revealing half-sentence was found in the "drafting." It confirmed, in line with the original assignment, that preliminary work by a specific graduate engineer would be considered. The chamber sees this as an indication of "fine-tuning" the input commands to the chatbot.

Overall, the explanations appeared to be a generic summary of the files, it continued. These also showed serious substantive deficiencies. For instance, the expert witness had apparently not even examined the plaintiff himself and only referred to an accident that had not actually occurred.

Fee Claim Disappears

Legally, the regional court based its decision on several paragraphs of the Judicial Remuneration and Compensation Act and the Code of Civil Procedure. A core point was the violation of the duty of personal performance. If an expert opinion is created to a significant extent by a machine or a third party, without the commissioned expert disclosing this and taking responsibility, the claim for remuneration is lost.

According to the decision, the expert witness vaguely stated upon inquiry that the overall responsibility remained with him. However, he could not dispel the doubts about his authorship.

In addition to the AI problem, the chamber criticizes the disproportion between the billed time spent and the actual outcome. Even if the suspected AI use were disregarded, at most four working hours would have been appropriate for the mere one and a half pages of substantive explanations. The remuneration for this should have been significantly lower than the invoiced amount.

The chamber thus emphasizes that, in principle, expert witnesses in the judiciary are allowed to use digital tools. However, they must at least mandatorily declare them and fulfill their mandate. Concealing AI-generated content, according to the decision, not only leads to the unsuitability of the requested evidence for the proceedings but also ultimately leaves the expert without payment.

(vbr)