EPO AI Patent Ruling Human Skill Remains Supreme
In a recent significant decision, the Board of Appeal of the European Patent Office (EPO) addressed for the first time the use of Artificial Intelligence, specifically ChatGPT, in arguments related to patent claim interpretation. The ruling emphasized that AI tools like ChatGPT cannot replace the crucial role of a skilled person in these matters.
The Patent Dispute A Brief Overview
The case involved Rieter CZ s.r.o. (Rieter), who was granted patent EP 3 118 356 on November 4, 2020, for a procedure related to rotor spinning machines. Following opposition proceedings initiated by Saurer Spinning Solutions GmbH & Co. KG (Saurer), the EPO Opposition Division upheld the patent in an amended form on April 18, 2023. Saurer appealed this decision to the EPO Board of Appeal (the Board), aiming for the patent's revocation.
Rieter defended its patent, arguing that certain features were not disclosed in prior art if the claim language was interpreted correctly. Notably, Rieter supported its interpretation by presenting responses from ChatGPT to questions about the patent's technical terms.
EPO Board Rules Against AI in Claim Interpretation
On May 14, 2025, the Board delivered its decision (T 1193/23), overturning the Opposition Division's findings and revoking the patent. The Board concluded that claim 1 lacked novelty over prior art, and Rieter's five auxiliary requests also failed on grounds of novelty or inventive step.
The most groundbreaking aspect of the decision was its commentary on using ChatGPT for claim interpretation. The Board deemed Rieter's ChatGPT evidence inadmissible because the full responses and the context of the prompts were not submitted in writing, only mentioned orally.
However, the Board provided crucial general guidance. It stated that ChatGPT’s answers were “irrelevant” because claim interpretation fundamentally relies on the understanding of a skilled person in the relevant technical field. The Board elaborated:
“The general increase in the spread and use of chatbots based on language models ('large language models') and/or 'artificial intelligence' alone does not justify the assumption that an answer received - which is based on training data unknown to the user and may also depend sensitively on the context and the exact formulation of the question(s) - necessarily undermines the expert's understanding of the respective technical field (at the relevant point of time)”.
The Board acknowledged that “suitable specialist literature” could be used to support how a skilled person might understand patent claim terms, but Rieter had not provided such evidence.
Why AI Cant Replace The Skilled Person Yet
The concept of a notional skilled person in patent law assumes access to all existing state-of-the-art knowledge. While LLMs, with their vast data access, seem attractive for quickly assessing information, their methodology presents challenges. Understanding how an LLM arrives at a conclusion and ensuring its assessment stays within the appropriate context is difficult.
This EPO decision clearly states that LLMs like ChatGPT cannot serve as a direct substitute for the skilled person's understanding. The Board highlighted key weaknesses:
- Lack of Transparency: The training data used by LLMs is often not visible.
- Sensitivity to Prompts: Responses can vary significantly based on the phrasing and context of the input questions.
While the second issue might be partially addressed by providing more detailed input information, the first is inherent to current LLM operations, making it hard to rely solely on AI-generated content for claim interpretation. Another significant challenge is aligning an LLM's knowledge with the state of the art at a patent's specific priority date, as LLMs typically cannot easily restrict their knowledge to a point in the past.
Consequently, the importance of human expert evidence, supported by “suitable specialist literature,” remains paramount in EPO proceedings and broader European patent litigation. Nevertheless, LLMs and other AI tools are likely to see increased use in assisting legal practitioners with tasks such as prior art searches, literature reviews, and even drafting submissions.
* All quotations from T 1193/23 in this article are based on machine translation.
