Protection for AI applications

There are many examples of AI applications. While ChatGPT is the best-known AI application, there are many others.

Examples of AI applications are:

• Using a neural network for determination or prediction of parameters
• Network architecture
• Big Data: data identification and processing with machine learning
• Monitoring and diagnostic systems supported by AI
• Simulations supported by AI, for example to increase efficiency or (medical) risk assessment

Are inventions using AI patentable?

Many inventions using AI are patentable – if they meet certain requirements. For AI inventions, analogous to computer-implemented inventions and software inventions, the following applies according to EPO Guidelines G-VII, 3.3.1:
Computational models and algorithms are per se of an abstract mathematical nature, irrespective of whether they can be "trained" based on training data. Hence, the guidance provided in G II, 3.3 generally applies also to such computational models and algorithms.

Therefore, computational models and algorithms are only patentable, if they have a technical effect.

"technical character" and inventions using AI

An "overall technical character" is the requirement for patent applications. However, it is not necessary for all features of the invention to be technical. Thus, if an AI application has an overall technical character, it could be protectable. Contribution to the technical character may be, inter alia:

• A technical purpose
• Claim features contributing to the technical character of the invention
• A technical implementation
• The use of technical means

Your benefit - our expertise in protecting for AI applications

Would you like to achieve protection for your AI application?

Our patent law firm has special expertise in this field. Dr. Malte Köllner, founder and partner of our law firm Köllner & Partner, represents the plaintiff, Stephen L. Thaler, in the DABUS proceedings for the naming of an inventor in an AI-generated invention.

We can advise and assist you in all aspects of protection for your AI application, including

- Assessment of your invention
- Patent applications, nationally and internationally
- Drafting and filing
- We represent you in oppositions and nullity actions
- We take over searches, monitoring and expert opinions
- We are experienced, highly qualified and multilingual patent attorneys

Please contact us for more information at: info@kollner.eu

FAQ - frequently asked questions

1) Can AI be patented?

Yes—but not “AI as such”. Generally, technical inventions in which AI is used to solve a technical problem are patentable (e.g., image/signal processing, medical technology, control systems, production processes, robotics, energy management, cybersecurity). Pure business methods or “automation using AI” without a technical contribution are often not eligible for protection.

2) When is an AI invention “technical” enough to qualify for a patent?

Typical strong arguments include:

• Connection to the physical world (sensors/actuators/hardware, image capture, measurement system)
• Integration into a technical process (machine or process control)
• Technical effect (e.g., reduced computational load/energy, lower latency, greater robustness, improved measurement/control quality)
• Specific measures such as data preprocessing, model/inference optimization, fault tolerance, and safety mechanisms

The more clearly the problem → solution → technical effect are described, the better.

3) What specifically can be protected in AI?

Often, it is not “the model” alone that is the key, but rather the entire system. The following, for example, may be eligible for protection:

• Inference methods (preprocessing, decision logic, post-processing)
• Training/adaptation (e.g., drift detection, self-supervised learning, federated learning that has a technical effect)
• Data pipeline with a technical effect (sensor fusion, artifact removal, compression)
• System architecture (edge/cloud distribution, accelerators, memory/runtime optimization)
• Robustness & Security (plausibility checks, fail-safe mechanisms, adversarial robustness)

4) I’m using a standard model (e.g., Transformer)—can I still get patent protection?

Often yes, if the core innovation does not lie in the well-known standard model, but rather, for example, in:

• domain-specific integration (sensor technology/industrial processes)
• efficient inference under hardware constraints (edge, power, real-time)
• specialized pre- and post-processing with a measurable technical effect
• a new control/regulation approach or quality assurance

“We use Model X” is rarely sufficient. What matters is the technical design and distinction.

5) How important is training data? Can data be patented?

Data as such is usually not the best subject matter for a patent. Often the following is more useful:

• Patent protection for methods of generating/labeling/synthesizing data (if technically justified)
• Patent protection for data processing with a technical effect
• In addition, know-how and trade secret protection is often crucial for data (access rules, documentation, contracts).

6) Patent or trade secret—which is better for AI?

That depends on how easily the solution can be replicated:

Patent: is advisable if the product is visible on the market or reverse engineering is possible; also important for attracting investors, and also for partnerships, and licensing models.

Trade secret: is strong when details remain internal (data, parameters, deployment details) and can be kept confidential at an organizational level.

In practice, a hybrid approach is often optimal: a patent on the technical core and trade secret protection for data / parameters.

7) What are typical pitfalls in AI patent applications?

Common causes of problems are:

• Insufficient focus on technical effects
• Description remains at the “algorithm or mathematics” level
• Lack of variants or alternatives and in consequence claims that are overly narrow
• Differences from the prior art were not determined carefully enough
• Bad Timing: Publication of the invention (e.g., paper, website, pitch deck) before the application is filed

8) What should I prepare before we examine or file an AI invention?

The following are helpful for a quick, reliable assessment:

• One or two pages outlining “Problem – Solution – Technical Effect”
• System diagram(s) (data flow, sensors/actuators, edge/cloud, hardware)
• Training/inference steps, parameter ranges, variants
• Measurement values/benchmarks or other evidence (latency, energy, error rate, robustness)
• What is “standard” and what is really new?