An architecture practice’s completed commissions leave behind more than a portfolio. Drawings, specifications, models and visualisations record decisions that can be useful long after the original project. Generative AI creates additional ways to analyse and reuse such material, including uses outside the practice’s control.
The European Commission’s consultation launched on 29 September makes that commercial question timely. It seeks views on technology’s effects on copyright, including the use of protected content in AI, and remains open until 03 November.[1] It is a consultation on potential future measures and does not introduce a new licensing entitlement or require architects to stop using AI.
The commercial implications depend on which material a supplier wants to use, the permissions attached to it and the rights the practice can grant. A restriction in a drawing’s footer may express the practice’s position, but it does not settle ownership or override contractual permissions and supplier terms.
The archive is not a single package of rights
The first obstacle is ownership and permitted use. A project archive may contain the practice’s original material alongside consultants’ contributions, licensed photographs, manufacturer information and client-provided documents. Contractual permissions and confidentiality obligations can differ across that collection. Its value as training material does not establish that the practice controls every right needed to provide it.
The RIBA’s professional guidance treats architects as both creators and users of AI, with questions around copyright in material supplied to or produced by systems.[2] The institutional perspective is useful, but a UK professional discussion should not be mistaken for the law governing an EU transaction. The actual rights assessment must follow the relevant jurisdiction and contracts.
Licensing depends on whether the practice can authorise the proposed use of the particular files. Permission to publish an image in a portfolio may not extend to model training. Similarly, permission to use a consultant’s design on one project may not cover incorporation into a reusable commercial system.
This can change the deal’s economics. An archive that appears large may contain relatively little material that can be licensed without additional permission, redaction or preparation. Conversely, a smaller, well-documented collection may be easier to use and assess. Counting files is a poor proxy for counting commercially available rights.
The assessment should also identify what is confidential independently of copyright. A file can raise client confidentiality or security issues even where a copyright question is straightforward. Removing a project name may not remove identifying information from geometry, metadata or associated documents.
An opt-out has a defined legal function
Article 4 of the EU’s 2019 Copyright Directive provides a text-and-data-mining exception subject to conditions, including lawful access and the absence of an appropriate reservation of rights. For content publicly available online, the provision identifies machine-readable means of reservation. Article 3 addresses a separate research exception.[3] It is inaccurate to reduce this framework to a universal rule that all training requires a licence, or that everything online is freely usable.
The practical question for a practice publishing work online is how its rights reservation is expressed and maintained. A notice understandable to a human reader and a signal that automated systems can process serve different purposes. The Commission’s technical consultation explains its work on machine-readable reservation protocols and how they relate to the general-purpose AI code.[4]
Maintaining a rights reservation becomes more complicated when work appears across a practice’s website and third-party platforms. The practice may control the technical signals on its own site while having limited control elsewhere. Records of the method used and when it was introduced can help document the position expressed on each channel.
A reservation should not be described as a technical guarantee that scraping will cease. Its legal effect, enforceability and recognition in a particular dispute are separate from whether every crawler respects it. The commercial objective is to express the practice’s position effectively and preserve evidence, rather than promise prevention that the mechanism cannot assure.
Provider transparency will not identify every reused drawing
The AI Act places copyright-policy and training-content-summary obligations on providers of general-purpose AI models.[5] Those obligations should not be transferred indiscriminately to every architectural practice that uses an AI application. A practice’s role in supplying, adapting or deploying a system needs its own assessment.
The Commission’s training-summary guidance also explains an important limit: the published summary is not intended to provide a file-by-file inventory.[6] A practice should therefore be cautious about assuming that a provider’s transparency document will establish whether a particular project drawing was used.
Where a practice supplies material directly, its agreement with the provider can identify the files, permitted purposes, retention period and treatment of resulting models. It gives the practice a record of what it authorised that a general public description of training sources cannot provide.
The consideration may be a payment, access to a specialist tool or another contractual benefit. Its value depends on the rights granted, including exclusivity, transfer and sublicensing provisions, and the material’s treatment if the service ends.
An attractive short-term fee can be poor compensation for an unrestricted, transferable license over reusable expertise. Equally, refusing every use may prevent the practice from obtaining a tool it would value. The judgement depends on the actual scope and benefit, rather than a presumption that all licensing is exploitation or all participation is progress.
What architecture practices can contribute to the debate
Collaborative project archives raise questions that architecture practices could bring to the consultation before 03 November. These include the difficulty of identifying rights across multiple contributors, the cost of clearing material and the contractual terms needed for a workable licensing agreement. Specific examples could show how those issues affect transactions in practice.
Archive permissions and software procurement can create conflicting positions. A practice may reserve rights in its published work while accepting application terms that permit training or reuse of uploaded files. The significance depends on what those terms authorise and whether the practice has permission to grant it, including under its agreements with clients and contributors.
The decision should preserve a distinction between technical reuse and professional reliance. A system can retrieve or synthesise a detail from an old commission without establishing that the detail is suitable for a new site, regulation or material. The practice still needs a professional review process before using it in a live project.
The consultation concerns possible future copyright measures, but the underlying commercial questions already arise when practices upload project material or negotiate an AI service. An archive’s value depends partly on the permissions attached to its contents and the uses a supplier seeks. Clear records can support those negotiations; they cannot establish whether an old design detail is suitable for a new commission.
Sources
[1] European Commission, consultation on technology and copyright, 29 September 2026
[2] RIBA, key points for architects on AI and copyright
[3] EUR-Lex, Directive EU 2019/790, Articles 3 and 4
[4] European Commission, stakeholder consultation on AI and copyright compliance
[5] EUR-Lex, AI Act consolidated text, Article 53
[6] European Commission, training-content-summary template guidance