AI and nuclear technologies are no longer advancing in isolation. AI is accelerating progress in nuclear reactor design and control, enabling faster prototyping, smarter simulations, and real-time optimisation without the need for shovels in the ground. Meanwhile, the news that a dedicated nuclear reactor on Three Mile Island will power Microsoft’s AI data centres highlights the growing energy demands of AI itself.
This convergence is creating a wave of patentable inventions. These range from reactor hardware to the AI systems behind them, including reactor design and simulation tools, and intelligent controllers managing nuclear processes in real time. But securing patent protection in this space isn’t always straightforward. At the European Patent Office (EPO), AI inventions must produce a technical effect going beyond their implementation on a computer in order to qualify for protection. Knowing how to navigate these legal requirements is key to building robust, enforceable patents.
The European patent framework for AI and simulations
To obtain patent protection in Europe for AI inventions, or indeed any computer-implemented invention, an assessment must be made under the well-established COMVIK approach, whereby only features that contribute to the technical character of the invention are considered for assessing inventive step, and ‘non-technical’ features are disregarded.
The EPO’s Enlarged Board of Appeal confirmed this approach in decision G1/19, which specifically addressed simulations. The Board held that a computer-implemented simulation can be patentable if it produces a technical effect that goes beyond the simulation’s implementation on a computer, and this can arise in a number of ways, particularly if there is a direct link to a technical purpose, such as through a link to physical reality.
This decision is directly relevant to AI and nuclear, where simulation and prediction are core to many inventions. Identifying and clearly articulating the necessary technical effect is key to securing protection.
AI in reactor design and simulation
One of the clearest intersections of AI and nuclear power lies in reactor design and simulation. AI tools, including generative models, are now used to rapidly prototype reactor geometries, optimise component layouts, and run complex safety and performance simulations. But are such uses of AI, essentially involving simulations, patentable?
According to G1/19, a simulation or design process is only patentable if it solves a technical problem by producing a technical effect that goes beyond merely running on a computer. Merely pursuing a patent for the technical principles underlying the simulated system or process is not, in itself, sufficient to establish patentability. In the context of AI in reactor design and simulation, the necessary technical effect may be established in one of several areas.
A straightforward way to establish a real-world impact underpinning a technical effect would be to define a manufacturing step following from the result of the simulation/design process (as confirmed in G1/19, reasons 134). This approach not only strengthens the patent claim in terms of validity, but also expands its potential in terms of enforcement scope. In particular, the patented process may be asserted in respect of manufactured components where they are ‘direct products’ of the patented process. The Unified Patent Court (UPC) has widened the opportunity for claimants to seek evidence of infringement of process claims through search and preservation orders, which can allow for the inspection of premises and the collection of digital evidence, in territories where previously this remedy was not commonly available or was more onerous to obtain. Further, for unitary patents (UP), it may also be more difficult for infringers to circumvent liability by splitting their activities across borders. For instance, where a design process takes place in one member state of the UPC and manufacturing in another, the combined activity may still amount to infringement of a unitary patent – an outcome that may not have been achievable for European patents without unitary effect.
Another route to establishing the necessary technical character in a simulation/design process would be to demonstrate that the resulting design is an improved technical product. Consider an AI model that optimises the fluid dynamics within a reactor’s primary coolant circuit. If this simulation produces a novel circuit layout that reduces turbulence and improves heat exchange efficiency, this physical improvement could indicate the solution to a technical problem, thereby meeting inventive step requirements under European practice.
Avoiding the broken technical chain fallacy
A possible pitfall when drafting patent applications for inventions in this area is the “broken technical chain fallacy”. This occurs when the purported technical effect depends on a user’s (subjective) reaction to information that is generated by the technical features of a claim, rather than on direct technical results (see T1670/07).
For example, if a simulation generates a set of performance constraints that an engineer must then interpret to design a coolant circuit, the link between the simulation and the final product is broken by human intervention and interpretation. Even if the simulation makes the engineer’s job easier, the technical chain is compromised, and patentability is undermined.
In such cases, it may be necessary to rely on an alternative route to patentability. One such option would be to rely on an implied technical use of the simulation. To establish this, it may be necessary to limit the claim scope to a specific and limited technical use. In the coolant circuit example, this could require using the performance constraints to generate at least an aspect of the design. Relying on this route can be challenging, particularly if, for example, the simulation merely verifies a pre-existing design, rather than creates a new one. To successfully rely on an implied technical use, the claims should be drafted to refer to the inputs, processes, and/or outputs associated with the limited technical use, and the technical use should be emphasised in the description. The more clearly the claims are anchored to physical reality, the stronger the prospects for overcoming the patentability hurdles.
Alternatively, it may be possible to rely on the improved efficiency of the simulation method, which could arise, for example, by reducing memory usage or processor cycles (leading to a better running computer). Here, the simulation may be deemed to solve a technical problem, even if the output is not directly tied to physical reality. In such cases, it is essential that the claims are drafted to refer to all of the simulation steps that, when executed, bring about the better running computer.
AI-based controllers in nuclear systems
AI-powered controllers for nuclear systems represent another highly promising area for innovation and patent protection. Here, the link to a technical effect can often be more direct and robust.
Simulations can be used to predict future states of a reactor and adjust system parameters in real-time in order to maintain safety, optimise efficiency, or manage complex processes. For such innovations, a patentable invention may lie in two areas: (1) the predictive simulation for forecasting a future reactor state; and (2) the control system that uses the simulation’s output to manage the reactor.
Consider a controller for managing the level of Xenon in a reactor in order to avoid Xenon poisoning, which can impair efficiency and produce unstable transients in a reactor. The controller monitors the state of the reactor (e.g. isotope levels, neutron flux etc.) and, based on an AI model, predicts a future rise in Xenon. In response, the controller takes pre-emptive corrective action using the available actuators (e.g. control rods, moderators, coolant flow etc.).
In such scenarios, both the simulation and the controller may individually, or jointly, support a patentable claim. For instance, the invention could lie in the AI simulation providing for a more accurate or faster prediction, or it could lie in the way the controller uses that prediction to act, for more effective intervention. A predictive simulation that improves the operation of the computer running the simulation may also, as mentioned, give rise to a patentable claim.
Notably, the EPO Guidelines (G-II, 3.3) state that the use of a mathematical method (like an AI simulation) to derive or predict the physical state of a real object from measurements of physical properties can give rise to a technical contribution, regardless of what use is made of the results of the derivation or prediction. Accordingly, there is a strong foundation for protecting predictive AI in control systems.
Nevertheless, for the strongest prospects of success when pursuing patent protection for AI-based controllers, a patent application should emphasise the link to physical reality wherever possible and, where applicable, the real-world impact that the controller has.
Closing remarks
As nuclear power reasserts itself as a key part of the energy mix – driven by the demands of AI infrastructure – AI is in turn accelerating innovation across nuclear technologies. This synergy is producing breakthroughs in reactor design, control, and simulation. Protecting these innovations, however, requires careful consideration of the nuances of European patent law. Patentability depends not just on novelty, but on a clear technical contribution – whether through improved physical outputs, tighter integration with real-world systems, or enhanced computational performance.
At Carpmaels & Ransford, our multi-disciplinary Energy and Tech teams are well-placed to advise on these challenges. We help innovators navigate the nuances of the law to secure robust, enforceable protection for AI- and nuclear-based inventions.