Generative AI technologies such as ChatGPT or Claude are not only a topic in the economy, but also increasingly in intellectual property protection. In particular, the question of how inventions relating to artificial intelligence can be patented is keeping companies and IP experts alike busy. The European Patent Office (EPO) has sent out clear signals in recent months – and has gradually adapted its practice.
No patent protection for artificial intelligence as an inventor
First, the sobering news: AI systems themselves cannot be named as inventors under the current legal situation. This was already established in 2020 in the “DABUS” decisions. Inventions must continue to originate from natural persons. The reason: only people can be rights holders under European patent law.
For companies, this means that even if a technical solution approach was essentially generated by an AI system, a human inventor must be formally named – ideally someone who provides specialist support during the development process or classifies the result in technical terms.
Technical character remains decisive
Patent applications that deal with AI or its application are generally eligible for protection at the EPO, provided they make technical contributions. Inventions in the following areas are currently in particular demand:
- Pattern recognition
- Language processing (NLP)
- Robotics and autonomous driving
- Optimization of industrial processes through machine learning
According to the EPO, it is important that the AI approach is not only presented in a mathematically abstract way, but in the context of a specific technical application. A neural network for improving image quality in medical devices, for example, may well be patentable – but a pure language model may not necessarily be.
EPO guidelines: clarity for AI innovations
In its Guidelines , the EPO has further specified its examination practice. The most important points:
- The term “AI model” is now explicitly included in several passages.
- It was clarified how training data, learning algorithms and decision rules are to be described in the patent application.
- The Office expects applicants to provide an “adequate description” of the AI system, especially in the case of self-learning methods.
For IP strategists, this means documenting technical details at an early stage, using terms such as “machine learning architecture”, “training pipeline” or “technical solution scenario” – and always focusing the disclosure on the specific technical effect.
Conclusion: AI is not a free pass – but also not an exclusion criterion
Even if AI itself cannot take on the role of inventor, inventions based on AI technology are still patentable – under clear conditions. The decisive factor is the clear presentation of the technical contribution, the avoidance of abstract formulations and close adherence to the EPO ‘s current examination guidelines.
Those who register here with strategic foresight can secure important protective positions – especially in a dynamically growing field of technology such as generative artificial intelligence.