UK AI copyright: which creative catalogue cannot answer who may use what, for which purpose?

A publisher may welcome an AI search tool but reject model training; a studio may license reference access while protecting performer likeness; an AI supplier may know where data came from without holding the rights needed for every use. The public debate often compresses those decisions into “AI versus copyright”, leaving businesses with no practical route forward. The UK’s 2024 consultation closed, and government published a report and impact assessment in March 2026, but policy analysis does not silently rewrite existing copyright law or a private licence. The first commercial opportunity is a one-catalogue, one-use-case rights-and-provenance review that identifies works, rightsholders, access, intended acts, contracts, evidence and unresolved permissions. This reading shows how a rights, licensing or creative-technology specialist can package that decision and build recurring clearance control without promising ownership, non-infringement, a future legal outcome or licensing revenue.

Current copyright law protects qualifying works and controls specified restricted acts, subject to licences and statutory exceptions. Whether an AI activity requires permission depends on the works, acts, access, purpose, contracts and jurisdiction; a government consultation or impact assessment does not itself grant a general training right.

The Copyright, Designs and Patents Act 1988 remains the legal foundation. The project team should avoid announcing a universal answer before mapping the actual copies, analyses, outputs and communications involved. Legal counsel decides contested interpretation; the operational specialist builds the evidence needed for that advice.

Training a general model, indexing a licensed archive, retrieving passages for a user and helping an employee draft new material create different technical acts and commercial purposes. Treating them as one “AI use” hides which works are touched, what is stored and which permission or exception is being relied upon.

Start with a use-case diagram: input source, ingestion, retention, model or service, prompt, retrieved context, output, user and onward distribution. A hosted supplier may perform some acts while the creative business performs others. Contracts and technical controls must match that allocation.

The first purchase should be a fixed-scope review of one catalogue and one proposed AI use. It should end with a rights map, provenance gaps, contract constraints, permitted and prohibited acts, questions for counsel, technical controls and a decision to proceed, license, redesign, negotiate or stop.

The scope might be one image archive used for semantic search, one voice catalogue considered for synthetic production or one text collection offered for model access. A narrow decision creates value faster than an enterprise “AI policy” that cannot answer what anyone may do tomorrow.

A defensible provenance record identifies the work or dataset, contributor and rightsholder, source, acquisition date, licence or assignment, territory, term, permitted uses, restrictions, model or supplier, transformations and output destination. Missing fields should be visible rather than replaced with an assumption of ownership.

Licensing becomes practical when the parties can define valuable content, permitted acts, excluded uses, access controls, attribution or reporting, payment logic, audit evidence, term and exit. A licence cannot price uncertainty that neither party is willing to describe or technically control.

Not every catalogue should be licensed for model training. Retrieval, evaluation, internal analysis or limited production tools may offer narrower starting points. The specialist can structure options; the rightsholder chooses the risk and value position with legal advice.

The supplier should explain what material it receives, where it is processed, whether it is retained or used for training, which subprocessors participate, how access and deletion work, what output evidence exists and how contract restrictions are enforced. Marketing assurances are not a substitute for a usable data-flow record.

Supplier evidence should be proportionate to the use. A low-risk internal tool and a model-development partnership require different diligence. Where technical facts are unavailable, the decision may be redesign or a controlled pilot rather than an assumed permission.

Provenance can shorten clearance, segment licensable catalogues, support differentiated terms, answer buyer diligence and reduce the cost of proving what a rightsholder controls. Its value comes from enabling faster decisions and credible offers, not from collecting metadata that no commercial workflow uses.

Prioritise catalogue segments with identifiable owners, clean versions and demand. Use provenance status as a portfolio filter: ready to license, ready after named fixes, restricted, disputed or unsuitable. This creates a commercial pipeline without pretending every asset has the same value.

A defensible recurring service maintains provenance, reviews new AI uses and suppliers, controls permissions, records licences, samples outputs and reopens decisions when technology, contracts or policy change. It earns renewal by keeping creative use authorised and commercially legible as the catalogue evolves.

The cadence can follow releases, acquisitions, new suppliers and licensing negotiations rather than a generic monthly report. The service should show open permissions, expiring terms, unresolved contributors and uses operating outside the approved design.

A catalogue digitisation, AI supplier procurement, licensing enquiry, archive acquisition, synthetic-content initiative, contributor challenge or policy change can create a real buying event. Outreach should offer a bounded rights decision, not claim that the prospect’s current AI use is unlawful.

Publishers, labels, studios, agencies, archives, image libraries and AI suppliers form a large but fragmented market. Partnerships with legal, technology and rights-management specialists can reduce trust friction. Selected outbound works when it references an observable catalogue or product event rather than generic AI anxiety.

Launch when the specialist can define catalogue and use-case scope, protect confidential rights data, distinguish operational mapping from legal advice and deliver a decision without predicting future legislation. The partner must provide rightsholder, contract, technical and commercial access.

The 1988 Act, Intellectual Property Office consultation and the government’s March 2026 report bound public claims. Proposed models and impact assessments remain policy evidence unless and until law changes. Do not publish invented licensing values, infringement rates or claims that provenance guarantees ownership.

GetFishNet can test whether your creative-rights expertise, catalogue access, first review and delivery capacity form a credible acquisition opportunity. The free eligibility test examines current acquisition pain points and development synergies without promising licences, clients, revenue or legal outcomes.

If one buyer has an expensive permission uncertainty and your method can resolve it, we can design a tailored multichannel route around that decision and test market response before expanding.

Test your eligibility