Software, AI and electronics disputes turn on the technical implementation described in the claims. Marketing terms such as ‘intelligent’, ‘connected’ or ‘AI-powered’ say little about infringement or patent validity. The legal analysis must connect the patent to identifiable hardware, code, data flows and operating conditions.
In this guide: Identify the technical subject matter · Map claims to the deployed architecture · Develop reliable testing and expert evidence · Review prior art, disclosure and amendments · Manage design-arounds and the wider IP position

Identify the technical subject matter
Under the EPC, exclusions and technical character require a structured analysis. The use of a computer does not by itself establish inventive step, while the presence of software or machine learning does not automatically exclude patentability. Identify the technical problem, the claimed solution and the features contributing to it. Distinguish an abstract model or business objective from a technical application or implementation supported by the disclosure.
Map claims to the deployed architecture
Prepare a system-level description of devices, servers, modules, interfaces and processing steps. Identify which entity performs each operation and where. A method claim may require more than proof that a product contains suitable hardware. Preserve the relevant source-code version, configuration, model and deployment records. Product names and user-interface screenshots can help identify the system, but often do not prove the hidden operation on which infringement depends.
Develop reliable testing and expert evidence
Use repeatable tests with documented inputs, settings, outputs and limitations. Separate observed behaviour from assumptions about internal implementation. For AI systems, distinguish training from inference and identify whether the claim concerns the model, the training process or a specific technical use. Confidential source code or architecture documents may require a targeted production request and protective regime. Avoid presenting a black-box test as conclusive where several internal methods could explain the same result.
Review prior art, disclosure and amendments
Relevant prior art may include patents, academic papers, standards documents, software releases and public technical documentation. Establish publication dates and the disclosed version. Examine whether the claimed technical effect is supported and whether the invention can be performed across its claimed scope. Amendments should be traced to the original application. A broad claim to an outcome without sufficient implementation detail may raise different issues from a narrowly defined technical mechanism.
Manage design-arounds and the wider IP position
A software update can change the infringement analysis quickly, but it must be assessed against the actual claims and any court order. Keep old and new versions distinct and record deployment dates. Review component licences, customer rights and indemnities without assuming they provide complete patent clearance. Copyright, trade secrets and open-source obligations are separate layers of protection and responsibility; a patent strategy should be coordinated with them while preserving the legal distinctions.
Four practical examples
These hypothetical examples explain typical decisions and evidence issues. They do not describe client matters or predict a court outcome.
1. A firmware update changes a control sequence
An electronics manufacturer modifies the timing of a claimed processing step. Engineers document the revised sequence and test it under normal operating conditions. Counsel reassesses the claims, equivalents and any existing order before deployment. The update is evaluated as a technical change, not merely a new version number.
2. An AI claim concerns training, not use
A service uses a trained model, while the asserted claim is directed to a particular training method. The team investigates who trained the model, which process was used and the relevant territorial facts. Offering an AI service is not assumed to prove performance of every claimed training step.
3. A cloud feature is distributed across providers
The accused functionality depends on a device, a cloud platform and a third-party service. Counsel maps the steps and control relationships before selecting defendants or alleging infringement. The architecture supports both the legal attribution analysis and any request for technical evidence.
4. A public repository may be prior art
A defendant identifies code said to predate the patent’s priority date. It verifies the accessible version, publication history and technical disclosure. A current repository page or an unexplained timestamp is not treated as sufficient proof of what the public could access at the relevant date.
Related guidance
Claim Construction and Equivalents · Product and Process Patents · Orders to Produce Evidence · Confidentiality and Court Records
Legal framework and sources
Articles 52–57, 83 and 123 EPC; UPC substantive infringement and evidence provisions.
EPO Guidelines: AI and machine learning · EPO Guidelines: mathematical methods and technical contribution · UPC Agreement
Discuss your patent matter
For an initial enquiry, identify the patent, the product or process and any known court or launch deadline. Please leave out confidential technical details until conflicts have been checked.