Main Menu

Copyright © 2026 Patentamt | Alle Rechte vorbehalten.

Logo

Artificial Intelligence and Patents

Can AI applications be patented – and can an AI actually invent anything? Both questions are currently keeping patent offices around the world busy. The answers are clearer than many people think

Artificial intelligence (AI) is changing a great deal in our lives – and it is also creating new ways of making inventions. AI systems analyze data, recognize patterns and develop solutions at a speed and on a scale that human thinking alone can hardly match.

This raises more and more questions that patent law has to answer – above all in two areas that are currently the subject of much debate: the right to be named as an inventor on the one hand, and the patentability of AI-assisted inventions on the other.

Can an AI be named as an inventor?

The answer is clear: no. Only natural persons – human beings of flesh and blood – can be named as inventors. This question has not been discussed in theory alone. The European Patent Office (EPO), the United Kingdom and the United States all decided it along the same lines after an AI called DABUS had been named as the inventor in two patent applications: in every case the applications were refused – because the right to be named as an inventor is reserved exclusively to natural persons.

That does not mean, however, that inventions developed with the help of AI are generally unpatentable. As long as a natural person is named as the inventor and the other requirements are met, there is nothing to prevent the use of AI on the way to a new invention.

Which classification applies to AI inventions?

There is no single patent classification for AI. The closest match is the G06N group of classifications. But because AI inventions have to solve a specific technical problem, they are usually classified in the field where that problem lies – medical technology, automotive engineering or energy supply, for example. Searches should always be run with suitable search terms – such as “neural network” or “machine learning” – in the relevant field of technology.

When is an AI application patentable?

The same basic rules for software inventions also apply to AI: in patent law, algorithms and machine learning models are regarded as mathematical methods. As such, they are excluded from patent protection – unless they are embedded in a technical environment and solve a specific technical problem. So the decisive question is not “Is AI involved?” but rather “Does it solve a technical problem?” 

An AI application that optimizes the state of charge of an electric vehicle and thereby increases its range, for example, does solve a technical problem – and is therefore patentable in principle. In this sense, AI is a general purpose technology – much like electricity or the internet. The technology as such cannot be protected, but its specific application to solve a particular technical problem can be.

Patentable

  • Cardiac monitoring device: a neural network for detecting irregular heartbeats in patient monitoring (EP 0 850 016 B2). The AI method solves a technical problem.

Not patentable

  • Share price forecasting: an AI application that predicts how share prices will develop in the future. This is an economic problem, not a technical one.
  • AI-based business model: “schemes, rules and methods for doing business” are excluded from patent protection – even when they are implemented with the help of AI.

A particular challenge: disclosure

AI inventions raise a question that is less complex for conventional inventions: how precisely does the training of an AI model have to be described? A patent application must disclose the invention in such a way that a person skilled in the art can carry it out on the basis of the application documents. For neural networks, this means that the training method, the data used and the relevant model parameters have to be described in sufficient detail. Anyone who remains too vague here risks having the application refused for insufficient disclosure.

This is what happened with a "method for determining cardiac output" (EP 1955228 A2), for example – the European patent application was refused. The reason: the training of the artificial neural network according to the invention could not be carried out because it had not been sufficiently disclosed.

  • A person in front of a computer, a thought bubble with a lightbulb, and a certificate roll

    More tips from our experts?

    Join the free webinars offered by our IP Academy. In the webinar “Protecting Software with Patents,” you will learn everything about patenting AI, machine learning and simulation methods directly from our speakers.

    Find out more