Generative AI tools such as ChatGPT, Claude, Gemini, Midjourney, Adobe Firefly and Suno now allow businesses to produce text, images, music, marketing copy and software code within seconds. What was until recently a largely academic question has become a matter of real commercial consequence: who owns the intellectual property in material created using artificial intelligence? The answer determines whether such content can be licensed, sold, enforced against competitors, or relied upon as a genuine business asset.
Copyright law has traditionally assumed that creative works originate from human authors. Generative AI challenges that assumption, producing sophisticated text, artwork, music and software with relatively limited human input. The United Kingdom is one of the few jurisdictions to have legislated specifically for computer-generated works, but the relevant provision was enacted decades before modern generative AI existed, and its application to today's technology raises questions the courts have yet to resolve.
In summary, ownership depends on the extent of genuine human creative input. Where a human meaningfully shapes the output, the ordinary rules of copyright apply. Where the AI effectively determines the content with little or no human contribution, UK law provides a narrower and materially weaker form of protection, one which the Government has already indicated it intends to abolish.
1. Is the work computer-assisted or computer-generated?
The first step is to determine which category the work falls into, since this dictates which set of rules applies.
Computer-assisted works
Where a human author contributes sufficient original creative input by directing the tone, structure, composition or imagery of the work, and then selecting, refining or editing the AI's output, the resulting work is properly regarded as one of ordinary human authorship. AI functions merely as a sophisticated creative tool, no different in principle from a word processor, a digital camera or graphic design software.
Although the Copyright, Designs and Patents Act 1988 ("CDPA") does not use the term "computer-assisted", it is a convenient way of describing works in which technology assists, rather than replaces, the human author.
The human is the author, and the work attracts standard copyright protection: life of the author plus 70 years, together with full moral rights.
Computer-generated works
By contrast, where AI determines the expressive elements of the work with little or no meaningful human creative contribution, the work may properly be characterised as computer-generated.
Under section 9(3) of CDPA, where a literary, dramatic, musical, and artistic (“LDMA”) work is computer-generated, the law nonetheless assigns an owner: the person who undertook the arrangements necessary for the work's creation. Section 178 of the same Act defines a "computer-generated" work as one produced by a computer in circumstances such that there is no human author.
The resulting protection is considerably narrower: copyright subsists for only 50 years from the end of the calendar year of creation, and no moral rights arise.
In practice, most commercial use of generative AI falls somewhere between these two extremes. The greater the human contribution to the expressive elements of the final work, the stronger the argument that it is one of ordinary human authorship rather than a computer-generated work under section 9(3).
The limited role of section 9(3)
The authorities suggest that section 9(3) has a correspondingly limited role. Courts generally seek to identify a human author wherever possible. If the developer or the user contributes sufficient creative expression, they are the author. If multiple contributors meet the requirements, joint authorship may apply.
Only where no human has originated the work's expressive elements does section 9(3) determine authorship. As AI systems become more autonomous, section 9(3) may become more relevant, but increasing AI autonomy also makes satisfying the originality requirement more difficult, leaving the provision both significant and legally uncertain.
2. Is the work original?
By contrast to entrepreneurial works (such as films, sound recordings, broadcasts, and published editions), which do not require originality to attract copyright protection and are protected based on the investment and organisational effort involved in their production, UK copyright protects only LDMA works that are “original”, meaning that they reflect the author's free and creative choices, a standard confirmed by the Court of Appeal as remaining good law following Brexit.
The originality paradox
This creates an evident tension: section 9(3) assumes a work may exist despite having no human author, yet UK copyright law now follows the "author's own intellectual creation". So, if a work truly has no human author, it is difficult to see whose free and creative choices it can be said to reflect.
The point has not been authoritatively resolved. Commentators have suggested several possible answers: that section 9(3) operates as an exception to the ordinary originality requirement; that originality should be assessed by reference to the deemed author under section 9(3); or that copyright should subsist wherever the output would have been regarded as sufficiently original had it been produced by a human author.
The last of these approaches is generally regarded as best reconciled with the wider structure of UK copyright law, since it avoids protecting trivial or wholly mechanical output while preserving Parliament's apparent intention that genuinely creative computer-generated works should attract copyright.
Assuming the work satisfies the originality requirement, the remaining question is one of authorship and ownership.
3. Who is the author?
As copyright initially vests in the author, the identity of the author therefore determines first ownership (subject to rules governing employment and assignment).
The author of LDMA works is the person who creates the work, meaning the individual from whom the protectable expression originates (contrary to the author of entrepreneurial works who is the person responsible for the production arrangements).
Therefore, the enquiry proceeds in two stages. First, ask whether the work has a human author. If it does, as a human has made a sufficient original creative contribution to an AI-generated work, identify that human author, and the ordinary principles of authorship apply. If it doesn’t, as no human author can properly be identified, then section 9(3) applies and authorship is attributed to the person making the necessary arrangements for the work's creation.
Computer-assisted works
Where a work is properly characterised as computer-assisted, authorship is determined according to the ordinary principles of copyright law. The central question is whether a human has made a sufficient original creative contribution to the expression embodied in the work. If so, that person is the author. AI does not displace authorship merely because it has been used in the creative process.
Brief, general prompts like "write a detective novel set in Victorian London" or "create a watercolour painting of a lighthouse at sunset" identify subject matter but leave language, composition and style to the AI, and are unlikely to confer authorship on the user.
Detailed prompting, iterative refinement, selection between outputs, and subsequent editing may, individually or cumulatively, demonstrate that the final expression originates substantially from the user rather than the AI.
AI developers will rarely qualify as authors of individual outputs. They design and train a general-purpose system capable of producing countless outputs, rather than determining the expressive content of any particular one, a position analogous to the developer of a word processor or design application. Only in the exceptional case where the developer retains direct creative control over the expression embodied in a particular output is authorship likely to arise.
Platform providers or the business deploying it are similarly unlikely to qualify merely by supplying the technology or the environment in which it is used, however commercially significant that contribution may be. Authorship depends upon the origin of the work's protectable expression rather than the provision or operation of the tools by which it is created.
UK courts have not yet considered authorship of generative AI output directly, and overseas approaches diverge: the US Copyright Office has generally taken a restrictive view of prompt-based authorship, whereas the Beijing Internet Court has recognised copyright where a user's prompts and iterative refinement demonstrated sufficient creative control over the resulting expression. The position under UK law is likely to depend on the facts of each case and upon the extent to which the protectable expression reflects the user's own intellectual creation.
Computer-generated works
Where no human can properly be identified as author, the enquiry shifts from who contributed the expression to who was responsible for organising the process by which the work came into existence.
The CDPA does not define "arrangements necessary", and there is no authority applying section 9(3) to generative AI. The expression is likely to be interpreted purposively, identifying the person most directly responsible for causing the particular work to be generated.
On this basis, the developer of the underlying model is unlikely to satisfy the test, having built a general-purpose system rather than arranged for the creation of any specific output. The platform provider, similarly, merely makes the system available and maintains its infrastructure.
The user who initiates the generation process is therefore the strongest candidate even where the prompt is insufficiently creative to establish ordinary authorship, it is the user's decision to employ the system and generate that particular output which most directly constitutes the arrangements necessary for its creation.
Employees
Where a work is created in the course of employment and int the absence of agreement to the contrary, first ownership vests in the employer, under section 11(2) of the CDPA, in the same manner as any other copyright work.
This applies equally to AI-assisted and computer-generated output, and employment contracts and IP policies should be updated accordingly, expressly addressing the use of generative AI, specifying which tools may be used, the purposes for which they may be used and confirming that copyright and other intellectual property rights in the resulting work belong to the employer.
4. AI providers' terms of service
Providers' terms of service are commercially significant, as they govern the contractual relationship between provider and user, but they do not determine whether copyright subsists under UK law or who is regarded as the author under the CDPA.
For this reason, most providers adopt carefully qualified language. OpenAI and Anthropic assign such rights as they hold in outputs "to the extent permitted by applicable law" and "if any", recognising that neither can assign rights it does not hold or create copyright where none exists.
Suno assigns its rights in generated songs to paying subscribers while restricting free-tier users to personal, non-commercial use. Adobe offers contractual indemnities to qualifying Firefly customers without warranting that copyright subsists in every output, and Midjourney has granted broader commercial rights to paid subscribers, though its terms have changed over time.
A contractual assignment cannot create copyright where none exists. Once copyright has vested in a first owner, it may of course be assigned or licensed in the ordinary way, an assignment must be in writing and signed by the assignor, while licences may generally be granted more informally. But this presupposes that copyright exists in the first place.
Businesses should examine precisely what is being assigned or warranted, rather than relying on the label attached to the provision.
5. The position is likely to change
In March 2026, the Government published its Report on Copyright and Artificial Intelligence, prepared under sections 135 and 136 of the Data (Use and Access) Act 2025, following a consultation that closed in February 2025 and attracted over 11,500 responses.
The Report indicates a clear intention to remove protection for purely computer-generated works under section 9(3), on the basis that there is insufficient evidence that the provision serves any meaningful purpose, while continuing to monitor its use in the meantime.
Works created with genuine human input are unaffected and will continue to receive standard copyright protection as they do today.
No timetable for legislation has yet been set, and the Government has indicated it will not act without a firmer evidential basis. The direction of travel is nevertheless sufficiently clear that businesses should not treat section 9(3) as a durable basis for protecting purely AI-generated output.
6. Practical recommendations
Maximise and document human creative input. Workflows in which a person makes meaningful, recorded creative choices like selecting, editing, arranging and refining AI output, place the resulting work on the firmer footing of ordinary, human-authored copyright, rather than the more fragile section 9(3) regime.
Treat section 9(3) as a fallback, not a foundation. Where it may be the only available protection, consider additional layers: confidentiality and trade secret protection, contractual restrictions with customers and partners, the UK's database right (where there has been substantial investment in compiling or curating data or outputs), and first-mover advantage.
Review AI providers' terms of service carefully, in particular what rights are actually assigned, what restrictions apply to commercial use, and whether any indemnity is offered in the event of a third-party claim.
Update employment contracts and IP policies to specify which AI tools are authorised, in what business contexts, and to confirm that resulting work is owned by the employer.
Take specific advice before relying on AI-generated output for anything of real commercial significance. Branding, packaging, key marketing assets and core software functionality are among the areas where the ownership gap presents the greatest risk.
This area of law remains unsettled and continues to develop. Businesses relying on AI-generated content for commercially significant purposes are advised to seek specific advice on their position.
How To Get In Contact
To find out more or if you require assistance with these matters, speak with our Intellectual Property Team on +44 (0)204 600 9907 or email info@culbertellis.com.
Accurate at the time of writing. This information is provided for general information purposes only and should not be relied upon as legal advice.
.png)
