Artificial intelligence is revolutionising branches of industry, from the analysis of chemical data for active substance identification to autonomous systems that trigger emergency braking manoeuvres. But without the right protection, start-ups and innovative companies’ risk having their groundbreaking technologies copied and their market position fizzled out.

The crucial point is: what exactly is patentable in AI?

Basically, there must be a technical solution that is new, inventive and commercially applicable. The key point is that the idea of an AI application cannot be protected. Instead, specific technical approaches and processes that enable a particular functionality are patentable.

Focus on the algorithm:

In concrete terms, it is often the AI algorithms themselves that are patentable:

• Algorithms that minimise error rates through adaptive learning, for example in medical diagnoses.
• This also includes neural networks that detect tumours in MRI or CT scans and thereby enable more accurate diagnoses.
• AI algorithms that optimise driving behaviour and traffic flows in autonomous systems.
• Hybrid methods that combine machine learning with rule-based systems significantly expand the possibilities for patentability.

The strategic challenges:

The dynamics of AI systems, which evolve through continuous learning, confront us with the question: How do you define the static core of the invention that is patentable?

In addition, patents require a detailed description (disclosure requirements). This is where it gets tricky, because training data is often business-critical and represents a competitive advantage, but the question arises as to whether and to what extent it must be disclosed.

Practical tips for patenting AI algorithms

Based on the challenges and requirements for patenting AI solutions, companies should consider the following aspects:

1. Focus on the technical effect (technical solution): Do not patent the abstract idea or business model of the AI application. Instead, focus on the specific technical processes and algorithms (e.g. pre-processing algorithms or deterministic steps) that enable new functionality or technically improve existing processes, such as increasing energy efficiency.
2. Definition of the static core in dynamic systems: Since many AI applications evolve through continuous learning, it is crucial to clearly define the static, patentable core of the invention and to delimit it in the patent claim. The challenge lies in making dynamic models legally tangible.
3. Strategic handling of disclosure requirements: Patents require a detailed description of the invention. Carefully consider which parts of the system (e.g. pre-processing algorithms) need to be disclosed and which business-critical information (such as training data) must not be disclosed. One approach may be to patent easily verifiable components.
4. Verifiability and black box issues: Since AI systems often function as a ‘black box’, making it difficult to prove patent infringement, you should check whether easily verifiable components or pre-processing algorithms can be patented to make it easier to prove infringement.
5. Securing market position and investor confidence: View patents not only as legal protection, but also as a strategic asset. Patents prevent imitations, secure exclusive access to technologies and minimise risk for investors, as they are an indicator of innovative capability.