The status of AI-generated works under UK CDPA s.9(3).
The provision
UK Copyright, Designs and Patents Act 1988 section 9(3) provides: "In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken." The provision was enacted in 1988, well before generative AI in any contemporary sense existed. Three competing applications have emerged in recent commentary and litigation.
Application 1 — the operator-as-author reading
The operator-as-author reading holds that the AI operator (the company running the model and producing the output) is the "person by whom the arrangements necessary for the creation of the work are undertaken", and therefore copyright in the AI-generated work vests in the operator. This reading is the most literal application of the statutory text, and it produces a workable allocation of rights to a human person (the operator) where the AI itself cannot hold copyright under English law.
Difficulty: the reading produces results that conflict with the framework of the rest of the Act. Section 9(3) was drafted on the assumption that the "necessary arrangements" — selecting source material, configuring inputs, refining outputs — required substantive human judgement that warranted authorship status. Modern generative AI requires only a prompt; the "necessary arrangements" in the contemporary sense are largely the model architecture and training, not the per-output prompt. Under operator-as-author, every AI-generated work in commercial circulation has copyright vested in the AI operator regardless of the prompter's contribution — a result that is hard to reconcile with the originality requirement that other parts of the Act apply to author-authored works.
Application 2 — the prompter-as-author reading
The prompter-as-author reading holds that the user submitting the prompt is the "person by whom the arrangements necessary for the creation of the work are undertaken". The prompter selects the desired output, refines through iteration, and decides whether to use the result; in this reading those choices constitute the "necessary arrangements".
Difficulty: a single sentence prompt that produces a sophisticated output cannot reasonably be said to involve substantive arrangements for the creation of the specific work. The prompter has no detailed specification of the result; the arrangements are made by the model and its training data. Under prompter-as-author, copyright vests in the person who typed three sentences and clicked a button — a result that strains the originality requirement and produces ownership of works the prompter did not in any meaningful sense create.
Application 3 — the no-author reading
The no-author reading holds that section 9(3) does not apply to contemporary generative AI outputs at all, because the "arrangements necessary for the creation of the work" are distributed across so many parties (model architects, training-data providers, prompters, operators) that no single human person made them. Under this reading, AI-generated works that do not meet the originality requirement of human-authored works fall outside copyright protection entirely — analogous to the position the US Copyright Office has taken under Thaler v. Perlmutter and the subsequent registration guidance.
Difficulty: the reading is the most analytically honest about the statutory text's mismatch with contemporary AI, but it produces a substantial legal vacuum. AI-generated content with no copyright protection is freely copyable by anyone, which removes commercial incentive for human-AI hybrid creative practice and creates anomalous results where a human's substantive contribution to a hybrid work is unprotected because the AI's contribution is also unprotected.
CIP's position — the human-contribution test
CIP's position is that none of the three applications is fully coherent, and that the framework should encourage adoption of a fourth approach: the human-contribution test. Under this approach, copyright vests in the human contributors to a work to the extent of their substantive creative contribution; AI-generated elements that do not reflect substantive human creative choice fall outside protection.
Operationally, this means a hybrid work — a screenplay where a human authored the structure and an AI generated dialogue from prompts — has copyright in the human-authored elements; the AI-generated dialogue elements are unprotected unless the human's selection-and-refinement process amounted to substantive creative contribution. The framework's Mixed-Rights Architecture provides the per-layer rights-declaration infrastructure that makes this allocation operationally possible: each layer of a work declares its own rights-holder and authorship-basis.
The position is consistent with the US Copyright Office's contemporary practice (registration permitted for human-authored elements, denied for AI-generated elements), with the EU AI Act's transparency requirements (which presuppose that AI-generated elements are identifiable as such), and with the framework's overall position that subsisting rights subsist in human creative contribution regardless of AI involvement.
CIP recommends that any future amendment of CDPA s.9(3) adopt the human-contribution test explicitly, replacing the 1988 wording with a contemporary formulation that accommodates generative AI without producing the anomalies of the three competing readings.