Mon 16th Nov 2026

Intellectual Property rights in a world of AI and I/O

Liz Mills explains the IP challenges when using AI to express an idea in computer code commercially.

With the increasing availability and functionality of artificial intelligence (AI) and machine learning (ML) models, more and more people have been able to use AI/ML models to transform their application ideas into a working computer code, a process known as “vibe coding” or “prompt engineering”.  

 

The input/output (I/O) process is simple: you put in a prompt that tells the AI/ML model what you want it to do, and the AI/ML model outputs a response and/or product based on your instructions and its training.

 

In fact, the practice of using AI to express an idea in computer code has become so accessible that Collin’s dictionary named “vibe coding” as their word of the year at the end of 2025!

 

However, although the practice of vibe coding has become more accessible, there are several potential Intellectual Property (IP) pitfalls that must be navigated if you would like to use your computer code commercially. The following provides a brief overview of some of these pitfalls, covering the different types of IP rights that are associated with vibe coding, whether those IP rights are available for computer code generated using vibe coding, and who actually owns any IP rights that are available.

 

Types of IP rights

Before we go any further, let’s briefly highlight what types of IP rights should be considered when vibe coding. The main three IP rights are: patent rights, copyrights, and trade secrets.

 

Patent rights

Patents are a form of legal protection for inventions, where an invention is a technical solution that has been made to address a technical problem (i.e., a problem that you need to overcome in implementing technology). Patents can be granted for inventions that are new, inventive, and capable of industrial application.

 

To get a patent, you have to file a patent application (i.e., a request for a patent to be granted). Patent applications contain a description of your invention, along with any accompanying Figures, and a set of claims. The claims define the legal scope of protection that you would like to cover with a patent, andcovers your technical solution. Historically, it has been difficult to obtain patent protection for computer-based inventions, with many jurisdictions specifying that you cannot get a patent for subject-matter that is a computer program “as-such”: it has to make a technical contribution to be eligible for patent protection.

 

Some important considerations for patent rights include:

  • In large part, you should not disclose your technical solution (e.g., make it public to friends, family, or to anyone else) before applying for a patent. If you do, this disclosure could mean that you're not allowed to be granted a patent.
  • A patent right is jurisdictional. This means, for example, that a UK patent cannot stop anybody from implementing your technical solution in another country.
  • A patent often takes several years to grant and can provide protection for up to 20 years from the filing date.
  • A patent is an absolute right and can be used to stop somebody else from performing your technical solution in the jurisdiction in which it is granted regardless of whether that person independently thought of the same technical solution.

 

Copyrights and Copylefts

Copyright is a form of legal protection that protects a particular expression of an idea. For example, a history textbook may differ from another history textbook on the same subject by the choice of language used, the ordering of the information, and by selection of which information to include and which to omit. The degree of copyright protection available can vary with the originality of an expressed idea, with more original work having a stronger copyright protection that less original work.

 

Some countries require that a copyright be registered with a Governmental body in order to be enforceable. However, this is not universal. For example, it is automatically granted in the UK, with different types of work being associated with different durations of protection.

 

Not every country allows for computer programs to be afforded copyright protection, although the UK does. In the UK’s case, copyright protection is available for computer programs for up to 50 years after its first written, as it is considered to be a “literary work”.

 

Copyright is not an absolute right: it protects against copying, but does not protect against independent creation.

 

Copyleft is a term that is less frequently used. The term Copyleft refers to potentially copyrightable matter that has been made available for other people to use without enforcement of copyright. This is an important concept in open source systems.

 

Trade secrets

A trade secret refers to any information that is kept confidential by a business for commercial advantage. Examples include formulas for food/drink, manufacturing methods, customer lists, etc.

 

No registration is required to protect a trade secret, but ongoing activities should be performed to maintain the secrecy. Trade secrets can potentially last forever and may be especially useful for covering things that are difficult for other people to discover by reverse engineering, such as prompts for AI/ML models.

 

Is an IP right available?

The question of whether a computer code generated by an AI/ML model can be the subject of a patent right and/or a copyright is highly fact-specific and varies greatly between different jurisdictions.

 

The following will discuss some of the current positions held by various jurisdictions. However, as this technology is still in the very early stages, these positions may change as the technology becomes more developed.

 

Patents

For patents, several urisdictions consider that the person who inputted the prompt into the AI/ML modelto be the “inventor” of the resultant output.

 

For example, the US Patent and Trademark Office (the USPTO) has issued guidance via the Federal register that categorically states that AI cannot be an inventor of a technical solution: it can only be an assistive tool. Moreover, they stated that the identification of an inventor is fact dependent and centred on the idea of “conception”, noting that “The question is whether the natural person possessed knowledge of all the limitations of the claimed invention such that it is so ‘‘clearly defined in the inventor’s mind that only ordinary skill would be necessary to reduce the invention to practice, without extensive research or experimentation.’’ Analysis of conception turns on the ability of an inventor to describe an invention with particularity”.

 

Further, previous guidance issued by the UK in 2022 indicated that although they believed that AI could only be used in an assistive manner for developing inventions at the moment (i.e., it is a tool for inventors rather than a generator of inventions), that they would keep this area of law under review to support AI innovation.

 

What this means in practice is that if you use an AI/ML model to write code that allows your technical solution to be implemented, you are likely to be considered the inventor of that technical solution in the US and the UK.

 

Another important consideration for patents and AI/ML models relates to disclosure.

 

As mentioned above, it is important that you don’t publicly disclose your technical solution before filing a patent application.  

 

However, what a court considers a public disclosure can be very different to what you consider to be a public disclosure. In particular, there are very real risks that prompts inputted into some types of AI/ML models are likely to be treated as a public disclosure by courts.

 

Further, as some AI/ML models use their previous results for further AI/ML model training, there is a risk that other users of that AI/ML model will be provided with outputs that are based on your work, which may curtail your future commercial opportunities. Also, depending on the timeline of events, disclosures of such outputs by those other users can prejudice the grant of a patent for your technical solution.

 

It is consequently important to ensure that any AI/ML models used in the development of your technical solutions are “closed” or “enterprise” systems that have terms and conditions guarding against such disclosures (as opposed to so-called “open” systems, which more commonly do not have such terms and conditions). It is also important to to ensure that inventors receive training on the identification and usage of such systems so that they can better curate the information that they input into an AI/ML model.

 

Although not explicitly explored, these disclosure considerations equally apply in respect of trade secrets.

 

Copyright

Compared to patent rights, different jurisdictions are more divided on how to determine the originator of copyrightable work. Further, given the diversity of industries affected by copyright protection and the accelerating pace of change in AI/ML usage, many jurisdictions have chosen to issue guidance notes instead of pressing forward with statutory change in this area.

 

For example, in the US, distinctions are drawn between work that is purely output by an AI/ML model and work output by an AI/ML model that has been further transformed by a user, or that has been formed based on a specific original work.

 

In relation to this, the US Copyright office has issued guidance noting that regardless of how detailed the inputted prompt is, the output of an AI/ML model reflects an expression that is characterised by the model itself, and not by the inputting user. As a result of this, and as a result of the US Copyright office further noting that copyright cannot be held by an AI/ML model, the output expression by the AI/ML model is considered to not be created by the inputter and not afforded copyright protection. As a reminder, if there is no copyright protection available for an output (and the idea behind it is not patented), then anyone that has access to that output can copy and use it.

 

In contrast, when an input includes a piece of original work (e.g., a painting) and the AI/ML model is asked to transform that painting based on specific input parameters, there may be some copyrightprotection available in the AI/ML model output that belongs to the creator of the original work (to the extent that the original work can be identified in the output).

 

Further, when someone modifies an output of an AI/ML model, this modification may result in that person being considered a creator of the modified output. This may allow the modifier to be allowed copyright in respect of the modified output.

 

Looking to the UK, the UK currently does not make these types of distinctions between forms of output from an AI/ML model. This is largely because the UK uses statutes that pre-date the rise of generative AI to classify AI/ML models as an assistive tool to the person inputting the prompts.

 

For example, the UK has a specific statutory provision for computer generated works (CGW), with a report published early in 2026 stating that “In the UK, copyright protection is available for any literary, dramatic, musical, or artistic work which is “generated by a computer in circumstances such that there is no human author”. This computer-generated works (CGWs) protection lasts for 50 years from the date of creation. In terms of copyright ownership for CGWs, the “author” of such a work is deemed to be the person “by whom the arrangements necessary for the creation of the work are undertaken”. In the case of a general purpose AI which generates output in response to a user prompt, the “author” will usually be the person who inputted the prompt”.

 

Further, when the output is determined to have an originality based on choices made by the inputter, this may be covered by other forms of copyright protection in the UK. Such works are referred to as “AI-assisted works”.

 

This means that, in the UK, if you define the prompt and information to be input into an AI/ML model, you’re likely to be considered the creator of the resulting output of the AI/ML model. As discussed above, this may make any output computer code available for Copyright protection.

 

 

Who owns the IP?

Now we know where IP rights can exist in the output of an AI/ML model, the next question is who owns that IP right. Only owners of an IP right (or their licensors) can enforce them.

 

Unfortunately, the owner of that IP right will be highly dependent on both the jurisdiction in which the IP was generated, and the status of the person providing the input to the AI/ML model. This is because different jurisdictions have different requirements for who owns the rights to apply for a patent.

 

In relation to the UK, the inventor of an invention is generally the owner of the right to apply for a patentfor it unless they are employed as an inventor for a company, or they have a position within their company that would lead to that company being considered the owner of that right to apply for a patent (e.g., if they are a company director). There may also be other contracts at play that specifies whether the right to apply for a patent belongs to someone other than the inventor (such as when a contractor invents something while working under a contact for services for a company).

 

The US takes a similar (albeit more inventor-favoured) approach, with the owner of the right to apply for a patent initially being the inventor, unless the inventor has been employed to invent that.

 

In practice, this can mean that if you work for a company and vibe code in your spare time, in some scenarios that company may own the right to apply for a patent for inventions expressed by your computer code. There may also be penalties involved if you disclose any IP associated with your computer code outside of your company before they’ve made steps to secure any IP rights.

 

The same may also be true for copyrighted works: if the work is generated in the course of your employment for a company, the owner of that copyright may be the company. Like patent rights, a distinction is made between employees (who have a “contract of service”) and contractors (who have a “contract for service”).

 

AI/ML models

Another important factor to take into consideration is the AI/ML model you use to generate your code.

 

There’s recently been a lot of discussion regarding the rights of IP rights holders whose work has been used to train those AI/ML models. In particular, different Governments have considered whether an unlicensed IP right that has been used for training the model has had their IP right infringed by the output of such an AI/ML model.

 

At the moment, no final decision has been made on this area in any jurisdiction. However, it is increasingly likely that IP infringement decisions in this area (and whether any defence to this is available) will rely heavily on how the output of the AI/ML model compares to the data used for training that AI/ML model. In other words, a determination whether or not an output is likely to infringe on the unlicensed training data will depend on how transformative the AI/ML model has been on that unlicensed data.

 

For example, AI/ML models that use training data and/or input data that is in direct competition with the output of those AI/ML models are more likely to be considered to infringe those IP rights than when the output is unlikely to be in direct competition with the training data and/or input data.

 

One example of this may be, for example, if you provide an AI/ML model with somebody else’s code, and ask the AI/ML model to write new computer code that has the same effect.

 

Further, some jurisdictions have indicated that where licenses are more readily available for an AI/ML model builder to use an IP right for training data, that use of these IP rights for training without paying the license fee are more likely to be actionable. This would lead to not only the AI/ML model ownervulnerable to being sued, but also anybody that is commercially using any products or computer code output by such models. Put simply, when you make or develop a product using an infringing AI/ML model, it’s possible that that product may also be considered to be an infringement.

 

However, the issue is still very much jurisdiction and fact-specific for a particular AI-ML model, with various jurisdictions indicating that certain actions are unlikely to be considered as an infringement for the purpose of AI/ML models (e.g., the UK sought to implement a copyright and database exception that allows text and data mining for any purposes).

 

Some best practice points

Although the best advice for minimising commercial risk associated with intellectual property rights is always to check with a patent attorney or an IP lawyer, there are lots of things you can do to mitigate this risk.

 

Some of these include:

  • Reviewing your employment contract or contract for services to determine to what extent you need to provide details of your invention and/or computer code to your company.
  • Selecting your AI/ML model carefully: If possible, check what type of data your AI/ML model is trained on (e.g., how similar is the training data to what you want to make), and whether the AI/ML model is trained based on unlicensed or licensed data. Also, even after you’ve finished your computer code, perform periodic checks to determine whether any IP rights issues have arisen for that AI/ML model based on the training data used. If possible, always opt for closed and/or enterprise AI/ML models.
  • Providing inventors with training on usage of AI/ML models in respect of both closed and open systems.
  • Keeping a contemporaneous record of how you developed your computer code, including:
Which AI/ML models were used (e.g., name, publisher, version number)
Which date(s) you used which models
What prompts were used to generate which bits of code
Whether you modified any outputted code
When filing any patent application and/or registering copyright, not listing any AI/ML model as an inventor or author (although some jurisdictions may require that you disclose when an AI/ML model has been used to develop the work).
Keeping up to date with authorship and inventorship discussions in the countries in which you'd like to operate your product.

 

For companies, it’s also important to keep track of what projects your employees and contractors are working on outside of their contracted work. This is not only for making sure that it does not infringe on any IP rights you own, but also to help protect your staff by making them more alert to potential IP pitfalls. This can be done by simply having regular meetings with staff, and by providing them with IP training and support.

 

Author

Our offices

London

+44 20 7831 7929

3rd Floor, 26 - 28 Bedford Row, London, WC1R 4HE, United Kingdom

Leeds

+44 20 7831 7929

Suite 3.0B, Platform, New Station Street, Leeds, LS1 4JB, United Kingdom

Exeter

+44 20 7831 7929

Generator Hub, The Gallery, Kings Wharf, The Quay, Exeter, EX2 4AN, United Kingdom

Munich

+49 89 4132 46460

Schubertstraße 7, 80336 München, Germany