Thanks to EIP, our AI Month sponsors, for this guest blog outlining the intersection of Artificial Intelligence (AI) and Intellectual Property (IP). 

Written by Felix Hall and Ben Maling.

Developing, training, and deploying an AI system can involve a huge investment of time and money. The goal is typically to convert that investment into profit and/or a competitive advantage.

That’s where Intellectual Property Rights (IPRs) can help, whether by offering a legal tool to prevent others benefitting from your investment, or by providing a licensable asset to facilitate revenue streams.

In this article, we set out how different forms of IP protection can be usefully applied to different facets of an AI system, and how an understanding of their interaction can help optimise your IP strategy.  


A patent is a time-limited (up to 20 years) right to prevent others from using your (new and non-obvious) invention in a given territory. Technical details of the invention must be described in a patent application, which is usually published 18 months after the patent application is submitted. Examples of things you might consider patenting in the context of AI include innovative training processes, inference processes, applications of AI to a new problem, user interfaces, and new hardware for use with AI.  

Patents can be a powerful form of IP protection. However, patents may not be the answer for all AI systems, nor for all aspects of a given AI system (see e.g. trade secrets section below). The extent to which patents can be granted for “computer-implemented inventions” such as those involved in AI, varies between jurisdictions, so it is important to have sound advice as to the prospects of success.

Trade secrets

A trade secret is confidential know-how or other information that is valuable to a business because it is secret, and for which reasonable steps have been taken to keep the information secret. If someone obtains a trade secret without the owner’s permission, then remedies can be sought in court, such as an injunction (i.e., an order not to disclose or use the information) and damages (i.e., an order to compensate for any loss).

Trade secrets can cover a wide range of information such as model architectures, model weights, computer code, and methodologies for training, running, or applying an AI model, as well as any other valuable information or know-how relating to an AI system, as long as it is kept secret. 

Trade secrets can therefore be used to protect aspects of AI systems that patents can’t protect, provided that aspect can be kept confidential (e.g. run on a private server or cloud). However, there are also risks, such as the risk of information being publicly leaked (either intentionally or unintentionally), in which case the information is no longer a trade secret. Understanding the risks and benefits is crucial for implementing an effective ‘patent vs. trade secret’ strategy.    


Unlike patents, copyright exists automatically when you create an original work, such as computer code, and allows you to prevent others from copying or distributing the work (or adaptations of the work) without your permission. Unlike trade secrets, the work need not be kept confidential for copyright to apply.

You can prevent others from copying your work (e.g. code), or you can license it for others to use. A particularly pertinent form of licensing for code is open-source licensing, in which source code is made available and free to be used by any party, though there can be conditions on this freedom.

This can have benefits such as increasing adoption and enabling friction-free collaboration between organisations, but also involves risk as you can’t take back an open source licence if you change of mind. Therefore, you should ensure that you are fully aware of the IP implications before open sourcing (or otherwise licensing) your code and that the decision to open source fits your business model. 

Database rights  

Database rights (also known as sui generis database rights) are available for companies based in the UK and EU, and automatically protect the contents of a database providing there has been a substantial investment in obtaining, verifying or presenting the data. The database right allows you to prevent others from extracting data from, or otherwise reutilising, a qualifying database. Unlike copyright, the data does not have to be original.

Often there is substantial investment in obtaining or verifying training data for AI models, in which case a database of training data may qualify for the database right. There is also the tantalising possibility that a set of model weights may be considered as a qualifying database, though this remains untested, at least in the UK courts. 

IP Strategy

In summary, different IP rights can be used to protect different aspects of an AI system, and given the variability of AI systems, there is no ‘one-size-fits-all’ approach. It is, therefore, crucial to consider these rights together and how they apply to your particular AI project and business model as a whole, to develop the right overall IP strategy for your situation. That’s why at EIP we started Codiphy: commercial lawyers and patent attorneys specialising in AI and working together to provide fully-connected IP advice to help you meet your commercial aims. If this sounds like something you could benefit from, feel free to get in touch.

Felix Hall
Ben Maling

Shona Wright

Shona covers all things editorial at TechSPARK. She publishes news articles, interviews and features about our fantastic tech and digital ecosystem, working with startups and scaleups to spread the word about the cool things they're up to. She also oversees TechSPARK's social media, sharing the latest updates on everything from investment news to green tech meetups and inspirational stories.