European Patent Office Boards of Appeal Issues Notable Decision on Machine Learning and Patentability

The European Patent Office (EPO) Boards of Appeal recently issued a decision in case T 1669/21, examining the patentability and sufficiency of disclosure in claims involving machine learning (ML). The case concerned an ML-based method for monitoring the wear of refractory linings in metallurgical melting vessels. The decision underscores the EPO’s high standards for sufficiency, especially for AI-based claims, and emphasizes key technical requirements for obtaining patent protection for ML inventions.

Background

The patent, owned by Refractory Intellectual Property GmbH & Co. KG, claimed a method that employed an ML model to analyze data—such as wall thickness, material composition, and temperature—to predict the wear on the lining of a melting vessel. Vesuvius Group S.A. filed an opposition, arguing that the patent failed to disclose necessary technical details for a skilled person to implement the invention without undue experimentation.

Key Requirements for Sufficiency of Disclosure

In reviewing the case, the Board identified several deficiencies in the patent, emphasizing that ML-based patents must provide detailed, reproducible information to meet sufficiency of disclosure under Article 83 EPC. Key points included:

1. Training Data: The Board highlighted that the ML model must be trained on representative and comprehensive data. Without specific guidance on where to collect relevant training data or how the computational model should be trained, the invention lacked sufficient detail to enable a skilled person to carry out the invention over the claimed scope without undue burden.
2. Algorithm and Model Details: The patent did not specify the architecture, structure, or learning algorithms used in the ML model. The Board noted that the skilled person was left to determine a calculation model suitable to address the problem, as no specific requirements were provided. For a skilled person to execute the invention, details on the ML model’s architecture, training processes, and any adjustments that were made to meet the specific application requirements should be included.
3. Parameter Choices and Model Adaptation: The Board was particularly concerned with the absence of guidance on how to select and adapt model parameters, which are often necessary for tailoring an ML model to specific industrial applications. In this case, no information was provided on which model parameters were relevant, or the criteria for selecting the appropriate parameters. This lack of direction posed a barrier to reproducing the invention reliably.
4. Validation and Testing of Model: Lastly, the Board found that the patent did not discuss validation methods or provide a single concrete embodiment that can be reproduced. Validation provides a measure of the model’s effectiveness in real-world applications, which the EPO regards as important for reproducibility in AI-based inventions.

Implications for AI and ML Patents

This case serves as a clear reminder that AI-related patents at the EPO must be drafted with a high level of specificity. Patentees should anticipate providing concrete details about model selection, training processes, parameter tuning, and validation methods. The decision emphasizes that broad claims covering abstract ML techniques without sufficient technical detail are unlikely to meet the requirements for patentability.

Commentary

This decision adds clarity for practitioners in drafting AI and ML-based claims, particularly for industrial applications. Patentees are advised to clearly describe the technical steps involved, including how training data is sourced and used, and to define specific algorithmic parameters that contribute to the invention’s utility. Avoiding vague or overly broad language and focusing on practical, reproducible aspects can strengthen AI patent applications.

For guidance on securing IP rights for ML and AI technologies, please contact Andrew Currier and Ashley Chu at info@pckip.com.

PCK IP is a leading North American intellectual property firm based in Canada, specializing in the preparation, prosecution, and strategic management of patent and trademark portfolios for global markets. Our team of seasoned patent agents, engineers, scientists, and IP lawyers supports clients across diverse technologies, with expertise in coordinating cross-border IP strategies. Contact us today to explore how we can help secure and manage your intellectual property.

The contents of this article are provided for general information purposes only and do not constitute legal or other professional advice of any kind.