Authorship, Copyright & IP

Start with the honest position: on the specifics of copyright and IP, defer to a qualified lawyer and the current law in your jurisdiction. That is not this course dodging a hard topic; it reflects the genuine state of things in 2026. The law governing AI output is unsettled, actively litigated, and different from one country to the next - what is true in the United States may not hold in the EU, the UK, or India, and rulings and guidance are changing year to year. Anyone who tells you there is a simple, universal answer is overselling.

What this lesson can do - and what a working designer actually needs - is give you an accurate map of the terrain and the vocabulary to think clearly, so you know which questions matter, where the risks sit, and when to pick up the phone to a lawyer rather than guess. Treat AI IP the way you already treat structural or contractual questions outside your competence: understand the shape of the problem well enough to act prudently and to know your limits, and bring in the specialist for the decisions that carry real exposure. Ignorance is a genuine liability here; so is false confidence.

The first unsettled area sits upstream of you: the data these models were trained on. Large image and text models learned from enormous corpora scraped from the internet, much of it copyrighted - photographs, drawings, articles, artworks - generally without the original creators' explicit permission or payment. Whether that training was lawful (for instance under 'fair use' in the US, or various text-and-data-mining exceptions elsewhere) is exactly what a wave of high-profile lawsuits by artists, publishers, and image libraries is fighting over, and the outcomes are not settled.

Why should a designer care about a fight between model-makers and rights-holders? Two practical reasons. First, downstream risk: if a model can reproduce a training image closely - a recognisable artwork, a trademarked form, a distinctive building - output that echoes it too nearly could expose you to an infringement claim, regardless of the training debate. Second, provenance uncertainty: because you usually cannot see what a given output was derived from, you cannot fully rule this out by inspection. The pragmatic responses: avoid prompting in the style of a specific living artist or brand for commercial work; be wary of output that looks strikingly like an existing work; and prefer tools whose training data is more clearly licensed (some vendors now market 'commercially safe' models trained on licensed or owned libraries, occasionally with indemnities - read those terms rather than assuming them).

Architecture adds its own wrinkle here. In many jurisdictions buildings themselves carry copyright as architectural works, so prompting for output 'in the style of' a specific famous building or a named living architect is doubly fraught - it risks both the training-data question and a more direct resemblance claim. There is also the reverse risk that some designers overlook: your own work, and your clients', can be what a model ingests. If you paste your unpublished drawings into a tool that trains on inputs, you may be feeding your originality into a system that later serves fragments of it to strangers - a concern that connects IP directly to the privacy and confidentiality duties of the next lesson. The safe instinct is symmetry: be as careful not to leak others' protected work into your output as you are not to leak your own protected work into the model.

The second question is whether AI output can be copyrighted at all, and it turns on a principle many jurisdictions share: copyright protects human authorship. Guidance in several countries - notably US Copyright Office positions through 2023-2025 - has held that output generated purely by an AI from a text prompt, with no meaningful human creative control, is not itself copyrightable, because a human did not author it. Purely machine-made output may simply have no copyright - meaning you cannot stop others from using it either.

The crucial nuance for assisted design is that human authorship is a spectrum, and this is where your own workflow matters. The more substantial, creative, and specific your human contribution - selecting, arranging, heavily editing, combining AI elements into a larger original work, using AI as one tool within a human-authored design - the stronger the argument that the overall work carries protectable human authorship, even if some components were AI-assisted. A raw one-prompt image sits at the weak end; a developed design in which AI played a supporting role sits at the strong end. This is yet another reason the whole course insists on AI-assisted, not AI-generated: keeping the human firmly in authorship is not only better design and better ethics, it is very likely better for your IP position too. Exactly where any given work falls, though, is a legal judgement - map it, then confirm with counsel.

Because the default legal position is uncertain, the practical protection is explicit agreement. Do not leave ownership of AI-assisted deliverables to be inferred later; address it in your contract, the way you already address ownership of drawings, models, and design rights. In most professional relationships the client expects to receive rights in the final deliverables they paid for, and your agreement should say clearly what is being transferred or licensed - and it cannot cleanly transfer rights that may not exist, which is precisely why the copyrightability question above matters to the wording.

Several clauses are worth discussing with your lawyer. Disclosure of AI use, so the client is not surprised and consents to the approach. Allocation of the IP in AI-assisted work between you and the client. Warranties and indemnities - be extremely cautious about warranting that AI-assisted output infringes no third-party rights when the training-data question is unresolved; you may not want to carry that risk, and you should understand what your tool vendors do and do not indemnify. And check the terms of service of every AI tool you use commercially: they vary enormously on who owns generated content, whether commercial use is permitted, and whether your inputs may be used to train the vendor's models (which ties directly into the confidentiality issues of the next lesson). The theme throughout: replace uncertainty with written clarity, drafted by someone qualified, before a dispute forces the question.

Beyond what the law strictly requires, there is the question of how you represent your work - a matter of professional ethics and client trust as much as copyright. Passing off heavily AI-generated work as wholly your own hand, or as another firm's or artist's style without acknowledgement, is a reputational and ethical hazard even where it is not clearly illegal. The reasonable posture is honesty proportionate to the context: you need not annotate every keystroke, but material reliance on AI in what you deliver or publish is something clients and the public increasingly expect to be disclosed, and the next lesson treats that disclosure duty directly.

There is also a forward-looking reason to keep your human authorship strong and documented: as norms and law settle, being able to show how you directed, selected, edited, and integrated AI - your genuine creative contribution - is what will best support both your ownership claims and your professional integrity. Keep a light record of your process on significant projects. In short: use AI openly, keep yourself unmistakably the author, respect others' rights and styles, be honest about your methods, and get the contracts right. Do those, and you are conducting yourself well within an unsettled landscape - but let a lawyer, not this lesson, write the words that bind you.

It is worth ending on perspective rather than anxiety. Every one of these questions - training-data lawfulness, copyrightability, ownership, attribution - is genuinely being worked out, and a working designer does not need to resolve them to practise responsibly; you need to recognise them, act prudently, and know when a decision has crossed from routine into the territory where specialist advice earns its fee. That judgement of your own limits is itself a professional skill, and it is the same one you already exercise around structural, contractual, or regulatory questions outside your competence. Treat AI IP no differently, and you neither freeze in caution nor blunder in overconfidence.