Does copyright subsist in AI-generated content?

What's the issue?

GenAI tools can rapidly produce trade mark logos, music for advertising, marketing copy, social media posts, adverts and packaging designs. This efficiency and creative power can be seductive, but an important question lurks: does copyright subsist in the outputs?

In short, the answer is that copyright probably does subsist in AI-generated outputs in the UK. However, there are some uncertainties in the legislation and there is a real risk that the government will limit protection in the future. In view of this – and the fact that many countries around the world do not offer copyright protection for AI-generated works - businesses should consider to what extent and how they rely on AI as part of their creative processes. 

Protection for computer-generated works

At present, there are two ways in which works could benefit from copyright protection in the UK where AI is used as part of the creative process:

• Computer-generated works. The Copyright, Designs and Patents Act 1988 (CDPA) provides a 50-year term of protection for computer-generated (literary, dramatic, musical and artistic) works where there is no human author (CGWs). It is not completely clear what the words "no human author" mean, but they could mean that there is no human originality. In the case of AI, this form of protection could arise where a simple prompt is used to generate an output. 

• Standard copyright protection. The CDPA also provides 'standard' copyright protection for original literary, dramatic, musical and artistic works. The term of protection is life of the author plus 70 years. These works require a human author/originality.

In practice, it is likely that a work will benefit from one or other form (or both forms) of protection where AI is used as part of the creative process. Either there will be a human author/originality (such that the work is protected under standard copyright) or there will be no human author/originality (such that the work is protected as a CGW). However, the two provisions are not straightforward (see below).

CGWs: uncertainties

The main uncertainty around CGW protection centres on the requirement for all literary, dramatic, musical and artistic works to be original and the fact that the current test for originality seems to require a human author. In particular, to be original, a work must be the author's 'own intellectual creation' and a reflection of their 'personality' and 'free and creative choices'. This leads to an odd result: CGWs by definition have no human author and yet to satisfy the originality test, they seem to require a human author. This tension has been widely acknowledged, even by the government. 

The best view seems to be that (if asked) a court would try to give effect to Parliament's intentions and confer copyright protection on CGWs. One way it could do this is by holding that a CGW is original if it would have been original had it been created by a human. However, it is also possible that a court would interpret the provisions differently (in ways beyond the scope of this article) and possibly even hold that CGW is not available in practice. 

We don't consider this in detail here not least because the debate might turn out to be academic. This is because the government has confirmed that it is likely proceed with its preferred option of abolishing copyright protection for CGWs (as set out in its Copyright and AI consultation). In this scenario, businesses would only be able to rely on 'standard' copyright protection, assuming they want copyright to subsist in their works.

Standard copyright protection: uncertainties 

There is also uncertainty as to how 'standard' copyright protection applies where AI is used as part of the creative process. 

• The government has said that standard copyright protection should be available for AI-assisted works, by which it seems to mean works created by a human author (ie with human originality) where the AI model is only used as a tool. The example given by the government of such a work is an original human photograph edited using AI. 

• Conversely, the government has said that standard copyright protection is not available for AI-generated works, by which it presumably means works created without human authorship/originality (ie works currently protected as CGWs). 

However, the government's view is obviously not binding on a court and there is no clear indication of where the line is between AI-assisted and AI-generated works. In particular, it is unclear whether using detailed, specific and creative human-authored prompts suffices for standard copyright to subsist in the resulting AI-generated output. While copyright might subsist in the prompt itself, opinion differs on whether it would/should subsist in the output. It might depend on the exact prompt used (and its level of 'originality') and the type of output in question. For example:

• There seems greater scope to argue that standard copyright subsists where a detailed and creative prompt is used to generate an artistic work as there would seem considerable scope for the human prompter to dictate what that artistic work should look like. 

• Conversely, where the output is a literary work, there seems less scope for copyright to subsist in the output since copyright only protects the precise form in which an idea is expressed/fixed, not the idea itself. Unless the prompt is very detailed as to what words to use (as opposed to the general idea that should be conveyed) in the output, then it is unlikely that a detailed prompt would suffice to confer the required human originality on the output. (We leave aside here any arguments about copyright in the prompt being infringed by copying the output.)

Unless the UK government chooses to clarify the position (unlikely), then these issues will only be thrashed out by court litigation. In the meantime, there is uncertainty as to how businesses can benefit from standard copyright protection while still incorporating AI into the creative process. This question will be particularly important if/when the UK abolishes protection for CGWs. 

Mixed works: uncertainties 

Lastly, it is also unclear how a court would treat a work partly consisting of a CGW (assuming the originality requirement is deemed to be fulfilled) and partly consisting of standard copyright. That could arise where AI is used to create an output (CGW for the output) which is then edited by a human author (standard copyright for the edits). In most scenarios, the same person would own both forms of copyright and would want to enforce copyright in the work as a whole (as opposed to the separate parts). However, it is not clear how the different terms of protection (50/70 years) and other issues would impact enforcement.

What about other countries?

Most countries (like those in the EU) only confer copyright protection on works demonstrating human originality. Whether copyright subsist when AI is used to help generate a work will therefore (as for standard copyright in the UK) be determined on a case-by-case basis. For the most part, copyright has been held not to subsist in AI works to date, as set out in our article here.

The US Copyright Office has expressly said that a prompt can never be sufficient to confer copyright protection on the output since outputs are significantly determined by the AI tool even where detailed prompts are used. We await case law to determine if this is correct. While the US Supreme Court recently declined to hear the Thaler case, that is not necessarily decisive and involves a slightly different issue around ownership. Nonetheless, for now, copyright can only be registered in the US where (and to the extent that) there is human originality. 

What does this mean for you? 

Given the current uncertainty, the possible approaches for the UK include the following:

• Accept some level of risk: for now, businesses might decide to incorporate AI into their processes in whatever way is commercially convenient on the basis that one or other form of copyright (or potentially both) is likely to subsist. While this would entail accepting some level of risk, it seems unlikely (although not impossible) that a work would be deemed to fall between the cracks such that it does not benefit from either form of protection.

• For key works and those with an international dimension: for those who require more certainty as to copyright subsistence, focusing on human authorship/originality and using AI merely as a tool to assist (such that standard copyright protection subsists) is the safest option. The added benefit of creating with 'standard' copyright protection in mind is that it maximises the chances of copyright subsisting in the work abroad since most countries require human originality for copyright to subsist. It should also mean that no/few changes are required to the creative process if/when the government abolishes copyright protection for CGWs. Planning ahead for that likely position is sensible now.

• Educate the business: make sure that the business is aware of the uncertainties and that your policies are fit for purpose. Mandate escalation if AI is used as the starting point (or is used extensively) to generate any works. 

• Consider how you want your outside agencies to use AI in the creative process: remember that, if an outside agency is also using AI to create works for you, then the same issues discussed above apply. It does not matter whether your agreements with your agencies state that all copyright is transferred to you – these are meaningless if no such copyright subsists. Changes to agreements with outside agencies to dictate when and how they use AI in the creative process might be required.

• Consider using only your own content as AI input: prompting AI using your proprietary material might not only reduce the risk of infringing third-party IP but also strengthen the argument that the output stems from your own creative ecosystem such that standard copyright protection should subsist. 

• Keep a detailed paper trail: record how you created each work in the same way that you would record how any other original work was created. This means saving your human drafts and prompt histories, screenshotting your iterations and logging your post-generation human edits. This audit trail might one day prove critical in demonstrating copyright subsistence and ownership.

• Exercise caution in communications: avoid overstating publicly that content is 'AI-generated' to preserve perceived copyright subsistence and ownership.

• Rely on other IP rights where possible: registering AI-generated logos and brands as trade marks should confer protection under the law of trade marks even if no copyright subsists in the work. Likewise, using the law of registered designs (where appropriate eg for surface decoration) should also confer protection. 

• Watch for developments: it is possible that the government will address some of the uncertainties by amending the CDPA.

• Make sure that you own any copyright that does subsist in CGWs: assuming copyright subsists in a particular CGW, a further issue is who owns that copyright. The CDPA provides that the author of a CGW shall be "taken to be the person by whom the arrangements necessary for the creation of the work are undertaken". However, there is no guidance on who that person might be (usually the human prompter but possibly not if a less detailed prompt is used). Fortunately, the standard T&Cs of most GenAI platforms provide that the user (licensee) owns all copyright in AI-generated outputs. Nonetheless, this should be checked on a case-by-case basis, including making sure that the appropriate type of licence is in place – see our article here.