Skip to main content
Back to Blog
Case Law·November 17, 2026·5 min read

BGH Decisions on AI and Software Patents H2 2026

Recent German Federal Court of Justice rulings on AI and software patentability - technical character, DPMA strategy, and comparison with EPO approach.

Dr. Anna Schmidt · Patent Attorney & AI Expert

BGH on AI and Software: How Germany's Highest Patent Court Is Recalibrating Patentability

The German Federal Court of Justice (BGH) published several decisions in the second half of 2026 that refine the understanding of software and AI patentability in Germany. For patent attorneys filing at the DPMA and EPO, these decisions are significant - not because they revolutionize the principles, but because they articulate the examination criteria with a clarity that previous decisions lacked.

The "Technical Character" Standard at the BGH

The central hurdle for software and AI patents in Germany remains the requirement of "technical character." The BGH has established a two-step test in its settled case law since "Logikverifikation" (X ZR 124/01) and "Sprachanalyseeinrichtung" (X ZB 20/17): first, the invention as a whole must exhibit technical character. Second, when examining inventive step, only the technical features may be considered - non-technical aspects (such as purely business rules or mathematical methods) are excluded.

In the second half of 2026, the BGH applied this test in several decisions, sharpening contours that are particularly relevant for AI practice. The core question remains: what makes an AI invention "technical"?

The September Decision: AI-Assisted Process Optimization

In a widely noted decision from September 2026, the BGH ruled on the patentability of a method for AI-assisted optimization of an industrial manufacturing process. The claimed method used a neural network to adjust process parameters in real time and minimize reject rates.

The BGH affirmed technical character with a remarkably clear rationale: a neural network is not to be regarded as an abstract mathematical model when it is embedded in a concrete technical control loop and acts on physical variables. The specific architecture of the neural network - including the choice of layers, activation functions, and loss function - qualifies as a technical feature when it was selected with regard to the concrete technical task.

This decision sends a clear signal: AI inventions embedded in a physical process have good prospects at the BGH. The difficulty lies with "pure" AI methods without physical interaction.

Pure AI Methods: The Boundary Remains Narrow

In a second decision from October 2026, the BGH examined a method for AI-assisted analysis of financial data. The system used a transformer-based algorithm for detecting trading patterns. The BGH denied technical character.

The reasoning follows the established line: the processing and analysis of data is only technical when it has a connection to a technical subject matter. Financial data is not technical data within the meaning of patent law, even if its processing occurs on a computer and employs an AI-specific architecture.

For practice, this means: the BGH draws the line not at the technology (AI) but at the subject matter. An identical AI architecture can be patentable when processing sensor data and unpatentable when processing financial data. What matters is not the "how" but the "what" and "what for."

Comparison with the EPO Approach

The BGH approach and the EPO approach (COMVIK method) reach the same result in most cases but differ methodologically. The EPO examines technical character in a unified approach within the inventive step analysis: non-technical features can define the closest prior art, but only technical features can contribute to inventive step.

The BGH separates more cleanly: first, technical character of the invention as a whole is examined (step 1), then only technical features are considered for inventive step (step 2). In practice, differences can arise, particularly for inventions that closely interweave technical and non-technical features.

An example: an AI method for optimizing a user interface (non-technical) while accounting for network latencies (technical) could be patentable at the EPO if the technical features go beyond the prior art. The BGH might have difficulty affirming technical character of the invention as a whole if the focus lies on UI optimization.

Implications for Drafting Strategy in Germany

The BGH case law from the second half of 2026 provides clear guidelines for formulating AI patent applications that will withstand DPMA examination and BGH review.

First: make the technical embedding explicit. Claims should not claim the AI component in isolation but embed it in the technical context. "Method for controlling a chemical reactor using a neural network" is stronger than "method for analyzing reaction data using a neural network."

Second: articulate the technical effect. The description must clearly set out what technical effect the AI solution achieves beyond mere data processing. Reduced reject rates, lower energy consumption, higher measurement accuracy - quantifiable technical improvements are key.

Third: justify the AI architecture technically. Why was this specific architecture chosen? If the answer is "because it delivers the best results for the technical task," that is a technical argument. If the answer is "because it is generally powerful," that is insufficient.

Key Takeaways for DPMA Filings

DPMA examination practice follows BGH case law but sometimes applies it more restrictively. The office published updated examination guidelines for AI inventions in autumn 2026 that reflect the BGH approach. Notable is the explicit recommendation to file AI applications with concrete working examples that demonstrate the technical effect - a point frequently neglected in software applications.

Patent attorneys should pay particular attention to the formulation of the technical problem in DPMA filings for AI inventions. A technically formulated problem ("improving the control accuracy of a process") is significantly preferable to an abstract one ("improving the prediction accuracy of a model").

Conclusion

The BGH did not trigger a revolution in the second half of 2026, but it set important orientation points. Patentability of AI inventions in Germany continues to hinge on technical character - and technical character hinges on embedding in a concrete technical context. Those who want to successfully file AI patents must present the invention not as an AI innovation but as a technical solution that employs AI as a tool. This perspective shift makes the difference between grant and rejection.


WunderIP helps formulate AI patent applications that meet BGH and DPMA requirements. Start your AI-assisted draft now.

BGHSoftware PatentsAI PatentsTechnical CharacterDPMA
Share

Ready to Transform Your IP Workflow?

Start your 14-day free trial. No credit card required.