AI innovation is moving fast—but can the patent system keep up? Drafting patents for AI systems isn’t just about protecting technology; it’s about bridging the gap between coders, technical experts, and legal frameworks.
With evolving regulations and the challenge of defining inventorship, securing strong IP protection requires careful strategy. Two IP patent attorney and AI expert Dr Coreena Brinck shares what you need to know.
Drafting patents for AI systems presents a complex array of challenges, particularly when it comes to inventions involving inventor teams who may be experts in different disciplines.
AI inventions can often be a result of collaborative efforts between coders, technical specialists in various areas, for example, medicine, and commercial managers, each with their own understanding of what the invention is, and possibly each having their own technical language to describe what may be the same thing in practice.
On top of this, are the various ways that code can be generated for any computer-implemented invention these days. For example, many coders and programmers rely on libraries of code snippets or pre-existing code modules and computer models to build their own platforms and systems and some also use generative AI applications to automatically generate source and/or executable programs using prompts (an example here being Claude, also available as a mobile phone app). This can lead to a lack of awareness over implementation and may bring additional challenges and complexity not just from describing how to implement an invention but also when it comes to identifying who exactly made various inventive contributions. The later is essential not just as the fundamental right to a granted patent rests with the inventor(s) unless assigned away, for example, to their employer, but also as only humans can currently have such ownership rights and this means non-human inventors are not allowed by patent offices.
Patenting complex technology accordingly means professional help may be crucial for success. Not only can a patent attorney help assess whether an invention is eligible for patent protection, but they can also determine who should be named as an inventor in view of their contribution to the invention. A patent attorney can also keep track of your patent application as it progresses which may include revisiting the question of inventorship as the inventors listed in the initial patent claims may differ from those who are ultimately listed in the granted claims, making the process even more fluid and challenging.
The issue of AI-assisted inventorship is further complicated as different patent offices have different guidelines and regulations. As of February 2025, the European Patent Office (EPO) does not allow non-human inventors to be named on patent applications. This rule means that AI systems, while they may play a significant role in the invention, cannot be listed as inventors, even though they may arguably have contributed to generating part or all of the inventive idea.
In patent drafting, a key challenge for protecting inventions involving applied AI, where a known AI model is used in a new way for a particular application, may be simply identifying what AI model(s) can be used in the invention and how their features and characteristics, including how the AI system is trained and functions operationally, contributes to a technical effect. The inventors may not have thought of their invention in this manner and it can be perplexing for them to understand what information they may need to share with their patent attorney so that the patent specification includes sufficient information to “teach” how their invention can be implemented by others skilled in that technical field.
Assessing what is sufficient information is not straight forward and can be particularly challenging for core-AI inventions. In some situations, during the training stage, details such as the volume, source, and cleaning/pre-processing of training data may need to be provided. What counts as “trained” and when the training process should stop may be other important factors to describe for some models. The weight learning algorithm of the ANN and initialization methods may need to be explained for some core models and it may be that a training method contributes to a different technical effect from that provided by the trained ANN model. The various possible training methods may also need to be described, and whether one or more different types of training are possible. For example, training may be supervised, unsupervised, on-going or continuous, in some situations.
If the training is on-going, it is crucial to describe what this means, as well as how errors are measured ongoingly and how this can impacts the performance of the model.
The operational stage of the AI system may be the core commercial proposition that the applicant wants to protect from its competitors. In this case, how the trained AI system is being used should be explained, and a causal link provided including any post-processing of its output so that the AI system as claimed provides a patentable technical effect.
Finally, given the ongoing evolution of AI technology and regulatory frameworks, considerations like the EU AI Act are become increasingly important. Inventors and patent professionals must navigate the fine line between patenting the AI system, protecting trade secrets, and utilizing open-source and defensive publications to safeguard innovations without over-disclosing sensitive details.
Ultimately, patent drafting in the AI space requires clear communication among diverse stakeholders with varying levels of expertise, clear and consistent language to be used in documentation of both the technical and operational stages, and some understanding of the AI regulatory landscape is also helpful. Successfully balancing technical disclosure, confidentiality, and patentability is what makes AI patent drafting a particularly complex and evolving field – as well as why I so enjoy working in this area.
If you are seeking guidance on protecting your AI inventions, you can contact Coreena via our website here or email us at hello@two-ip.com