Software and Artificial Intelligence Patent Guide

1. Can Software Be Patented in Europe?

Article 52(2)(c) EPC identifies programs for computers among the subject matter excluded from patentability. Under Article 52(3) EPC, however, this exclusion applies only to such subject matter or activities as such.

Consequently, the EPC does not impose a general prohibition on patents involving software.

A computer-implemented invention may be patentable when it satisfies the applicable requirements, including novelty, inventive step, industrial applicability and sufficient disclosure.

A claim directed to a computer-implemented method involving technical means will generally possess technical character. Nevertheless, the presence of technical means alone does not establish an inventive step.

The EPO distinguishes between:

• Technical features, which may contribute to inventive step.
• Non-technical features, which do not support inventive step unless they contribute to the technical character of the invention through interaction with other features.

This distinction is central to European software patent practice. For related service work, see our AI and software patents page.

2. The COMVIK Approach: How Inventive Step Is Assessed

The EPO applies the approach developed in decision T 641/00 (COMVIK) when assessing inventive step for inventions containing both technical and non-technical features.

Under this approach, features contributing to the technical character of an invention are considered in assessing inventive step.

Non-technical requirements may form part of the framework given to the skilled person, but they do not independently establish an inventive technical contribution.

For example, implementing a new commercial pricing rule on an ordinary computer will not necessarily involve an inventive step merely because the rule is new.

By contrast, a software implementation that improves the internal functioning of a computer, controls a technical process or achieves another demonstrable technical effect may provide a basis for patentability.

Practical consequence for patent drafting

A patent application should not merely describe what the software calculates.

It should explain:

1. The technical problem being addressed.
2. The technical means used to solve that problem.
3. The relationship between the claimed features and the resulting technical effect.
4. Why the technical solution is not obvious in view of the prior art.

The technical contribution should be identifiable from the claims and supported by the description.

3. Are Artificial Intelligence and Machine Learning Inventions Patentable?

Artificial intelligence and machine learning technologies are generally treated as computational models and algorithms.

Their mathematical nature does not automatically prevent patent protection, but neither does the use of AI automatically establish patentability.

An AI-related invention may make a technical contribution through its application to a technical purpose or through a technically motivated implementation.

Examples may include:

Industrial control

A machine-learning model processes sensor signals and adjusts operating parameters of industrial equipment to improve physical performance.

Medical imaging

An AI-based image-processing method identifies or enhances technically relevant image features to support a medical imaging process.

Wireless communications

A trained model is used to estimate channel conditions, reduce interference or adapt transmission parameters in a communications system.

Computer architecture

A particular neural-network implementation reduces memory accesses, processor workload or computational latency through specific technical measures.

In each case, the patentability assessment depends on the actual claimed features and their technical contribution.

Merely stating that artificial intelligence improves accuracy, efficiency or decision-making is generally insufficient without identifying the relevant technical context.

4. Technical Effect: The Central Question

For software and AI inventions, a critical question is whether the claimed features produce a technical effect beyond the ordinary execution of a computer program.

Technical effects may arise from, for example:

• Controlling a physical device or industrial process.
• Processing signals or images for a technical purpose.
• Improving the internal operation of a computer or network.
• Reducing computational resources through a specific technical implementation.
• Improving the reliability or performance of a technical system.

However, an improvement in the mathematical accuracy of a model does not necessarily constitute a technical effect.

Similarly, faster execution or reduced resource consumption must be assessed in relation to the actual technical implementation and the features responsible for the improvement.

Example: Neural-network classification

Consider an invention that classifies data using a neural network.

If the classification concerns purely commercial information, the improved classification accuracy may not, by itself, establish a technical contribution.

If the classification is specifically adapted to identifying a technical condition in sensor measurements, the claimed features may contribute to a technical solution.

The difference lies not simply in the algorithm, but in the technical purpose and the claimed implementation.

5. Disclosure and Enablement Requirements for AI Inventions

Article 83 EPC requires a European patent application to disclose the invention sufficiently clearly and completely for it to be carried out by a person skilled in the art.

For AI inventions, this requirement deserves particular attention.

A patent application should provide enough technical information to enable the skilled person to reproduce the claimed invention without undue burden.

Depending on the invention, relevant information may include:

• The model architecture and functional relationships between its components.
• The nature and technical characteristics of the input data.
• Training procedures and relevant parameters.
• Data preprocessing and feature extraction.
• The relationship between model outputs and technical operations.
• Implementation details necessary to achieve the claimed technical effect.
• Experimental results or examples supporting the disclosed technical contribution.

It is not always necessary to disclose every training example, model weight or source-code instruction.

However, omitting information essential to reproducing the claimed technical solution may create difficulties under Article 83 EPC.

The black-box problem

Describing an AI model merely as a component that receives inputs and generates desired outputs can be insufficient.

Where the claimed technical effect depends on particular training conditions, model structures or data characteristics, those elements may need to be disclosed.

The appropriate level of detail depends on the technical field, the knowledge of the skilled person and the breadth of the claims.

6. Claim Drafting Strategies for Software and AI

A software or AI invention may be protected through different claim categories, provided that each category is supported by the application and complies with the EPC.

Potential categories include:

Computer-implemented method claims

These define a sequence of technical operations performed by a computer or another technical system.

System or apparatus claims

These define technical components configured to perform the claimed operations.

Computer program claims

These may protect a computer program comprising instructions that cause a computer to perform a patentable method when executed.

Computer-readable storage medium claims

These may cover a storage medium containing instructions for carrying out the claimed method.

The most appropriate combination depends on the invention and the intended commercial protection. See also our patent drafting and prosecution approach.

Avoiding unnecessary limitations

An independent claim should ordinarily include the features necessary to define the technical solution without introducing implementation details that are not essential.

For example, specifying a particular neural-network architecture, programming language or hardware platform may unnecessarily narrow protection if the technical contribution does not depend on that limitation.

Conversely, removing essential technical features may weaken the inventive-step argument or create clarity and support problems.

A robust claim strategy balances breadth, technical contribution and adequate disclosure.

7. Protecting AI Innovations Through Patents and Trade Secrets

Patent protection is not always the only appropriate form of protection for software and AI innovations.

Certain elements may be better protected as confidential know-how or trade secrets.

Potential examples include:

• Proprietary training datasets.
• Internal model optimisation procedures.
• Non-public parameter configurations.
• Confidential data-processing pipelines.
• Commercial deployment methods that are difficult to reverse-engineer.

Patent protection generally requires public disclosure of the invention in exchange for a limited exclusive right.

Trade-secret protection, by contrast, depends on maintaining confidentiality and satisfying the applicable legal requirements.

A combined strategy may be appropriate where the patent application protects the technical invention while commercially valuable implementation details remain confidential, provided that the patent application still satisfies the applicable disclosure requirements.

8. Common Problems in Software and AI Patent Applications

Several recurring issues can weaken applications before the EPO.

Describing the business benefit instead of the technical effect

Statements concerning improved revenue, better recommendations or more efficient business decisions do not necessarily establish a technical contribution.

Claiming the desired result without defining the technical solution

A claim that merely requires a model to achieve improved performance may lack the technical features needed to distinguish the invention from the prior art.

Insufficient disclosure of the AI implementation

Where the technical effect depends on specific training or implementation details, a generic reference to machine learning may be inadequate.

Overly narrow independent claims

Unnecessary restrictions to particular models, datasets or software architectures may reduce the commercial value of protection.

Unsupported generalisation

An application demonstrating an effect for one specific implementation may not necessarily support a claim extending to substantially different implementations.

These issues should ideally be addressed during drafting rather than after an adverse examination report has been issued.

9. Preparing a Software or AI Patent Application

Before instructing a patent attorney, an applicant should ideally prepare a technical invention disclosure addressing the following questions:

1. What technical problem does the invention solve?
2. How is that problem addressed in existing systems?
3. Which technical features distinguish the invention from those systems?
4. What technical effect results from those distinguishing features?
5. What evidence, simulations or experimental results support the effect?
6. Which features are essential, and which represent optional implementations?
7. Has any relevant information already been publicly disclosed?

For AI inventions, the disclosure should also explain how the model is trained, implemented and integrated into the relevant technical system, to the extent necessary to understand and reproduce the invention.

Early identification of the technical contribution can significantly improve the quality of the patent specification and the resulting claim strategy.

If you are preparing a first filing more generally, see our First Patent Application Guide.

10. European Patent Strategy for Software and AI Companies

For companies developing software or artificial intelligence technologies, patent strategy should be integrated into the broader product-development and commercialisation process.

Important considerations include the technical maturity of the invention, potential public disclosures, intended markets, competitive alternatives and the expected costs of prosecution.

A European patent application may be pursued directly before the EPO or through the European regional phase of a PCT application.

The selection of the filing route should be considered separately from the substantive question of patentability. See our European patent, PCT and national filing guide and patent strategy services.

How Stan Advoka Can Assist

Stan Advoka advises on the preparation and prosecution of European patent applications involving software, artificial intelligence and other computer-implemented technologies.

Our approach focuses on identifying the actual technical contribution, developing a defensible claim structure and aligning the patent application with the applicant’s commercial objectives.

Depending on the invention, this may involve:

• Technical patentability and prior-art assessments.
• Evaluation of inventive step under the COMVIK approach.
• Drafting of descriptions and claims.
• Development of alternative claim categories.
• Assessment of disclosure sufficiency.
• Responses to EPO examination objections.
• Coordination of European and international filing strategies.
• Evaluation of complementary patent and trade-secret protection.

Applicants are encouraged to seek advice before publishing technical details or disclosing their inventions without appropriate confidentiality arrangements.

Contact Stan Advoka to discuss a software or AI patent matter.