Are machine-learning models patentable at the EPO?

A model applied to a technical purpose can be. A model claimed purely as a mathematical method, or applied only to a business purpose, generally is not.

Can an AI system be named as inventor?

Under current European practice the inventor must be a natural person. We document human contribution to the inventive concept accordingly.

Should training data be described in the application?

Where the data handling is part of the technical contribution, yes — and in enough detail to support the claimed scope.

Practice areas · AI

Algorithms, AI & Machine Learning

Models, training methods, data pipelines, AI-assisted inventions and inventorship questions.

Overview

Machine-learning inventions sit exactly where European practice is most demanding: a mathematical method is excluded, but its technical application is not. We work through model architecture, training regime, data handling and deployment to identify where the technical contribution actually lies, and we draft so that the contribution survives examination. We also advise on inventorship where AI systems contribute to the inventive concept.

What we do

Claim strategy for model architectures and training methods
Protection of data pipelines, labelling and augmentation methods
Technical-application framing for excluded mathematical methods
Inventorship and ownership analysis for AI-assisted inventions
Portfolio mapping across model, training and inference layers
Trade-secret vs patent analysis for weights and hyperparameters

Questions

Have a matter to discuss?

Tell us what you are building. We will run a conflict check and arrange an NDA before any confidential disclosure.

Discuss Your Matter All practice areas