The European Patent Office (EPO) Boards of Appeal (BoA) have issued an important decision addressing the growing intersection of artificial intelligence and patent law. In T 1193/23, decided on April 15, 2025, the Board established crucial precedent regarding the use of AI-generated content. This decision relates specifically to ChatGPT responses used as evidence of how a person skilled in the art would interpret a technical term in a patent claim.
Case Background
The appeal concerned European Patent EP 3118356, owned by Rieter CZ s.r.o., relating to safety mechanisms for rotor spinning machines used in textile production. The patent claimed a method for safely starting and stopping rotors by checking bearing regulation and data connections before allowing operation.
During oral proceedings before the BoA, the patent proprietor made a notable first: they verbally submitted responses from ChatGPT to support their interpretation of technical terms such as "bearing regulation." This marked the first time the EPO Boards of Appeal directly addressed AI-generated content as evidence for claim interpretation.
The Board's Ruling on AI Evidence
The BoA delivered a clear message about the limitations of Large Language Models (LLMs) in patent proceedings. In what has become the decision's key holding, the Board stated:
"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 a received answer - which is based on training data unknown to the user and can also depend sensitively on the context and the exact formulation of the question(s) - necessarily correctly reflects the expert's understanding of the respective technical field (at the relevant time)."
The Board identified three fundamental problems with relying on LLM outputs in patent contexts:
- Unknown Training Data: LLMs rely on training data that is unknown to users, making it impossible to verify the sources or quality of information
- Context Sensitivity: AI responses depend heavily on how questions are formulated and the specific context provided
- Temporal Accuracy: LLMs cannot be assumed to reflect actual expert understanding at the relevant priority date
Practical Implications for Patent Practice
What the Decision Does NOT Say
Importantly, the Board did not establish an absolute prohibition on AI-generated evidence. The decision makes clear that LLM outputs are not categorically inadmissible in patent proceedings.
What IS Required
The Board established that any AI-generated evidence must be substantiated with additional supporting evidence. As the Board explained: "Evidence of how certain terms in the claim of a patent (or patent application) are interpreted by the person skilled in the art can be provided, for example, by appropriate technical literature."
Best Practices Moving Forward
For patent practitioners considering AI tools in claim interpretation:
- Provide Context: When using AI tools, provide comprehensive technical context, including relevant case law and guidelines
- Document Everything: Maintain detailed records of prompts, responses, and methodology
- Corroborate with Traditional Evidence: Always support AI insights with technical literature, expert testimony, or established sources
- Consider Timing: Be mindful of knowledge cutoff dates for AI models when dealing with rapidly evolving technical fields
The Broader Patent Decision
Beyond the AI aspects, the Board revoked the patent for lack of novelty over prior art document D3. The Board found that the disputed technical terms should be interpreted more broadly than the patent proprietor argued, ultimately concluding that all claim features were disclosed in the prior art.
Conclusion
T 1193/23 establishes important guardrails for AI use in patent proceedings while acknowledging the technology's potential value. For patent practitioners, the message is clear: embrace AI as a powerful tool, but always within the established framework of substantiated evidence and technical expertise.
As AI technology continues to evolve, we can expect further refinements to these principles.
Patent professionals should stay informed about developments in this space while maintaining rigorous standards for evidence and technical analysis.