Ethical Use of AI in Creating Art

The Resistance cluster argued about whether AI in art is legitimate at all. The Reflection cluster reframed the question. The Practical cluster opens here, with the question every working artist who has decided to use AI now needs a precise answer to: what does it look like to do this *well*, on terms that earn the trust of audiences, collectors, clients, and the artist's own future self? This is the artist-facing ethics of using AI to create work. The training-side ethics — the rights of the artists whose work AI was built on — is the next article.

by Airtistic.ai Editorial

The Resistance cluster of this series argued, across four articles, about whether AI in art is legitimate at all — creatively, economically, legally, emotionally. The Reflection cluster reframed the question and named the configuration (AI-augmented human art) where the most interesting working practice is happening.

The Practical cluster opens here, with the question every working artist who has decided to use AI now needs a precise answer to:

What does it look like to use AI ethically when creating art?

This article addresses the artist-facing side of that ethics — what working artists owe their audiences, their collectors, their clients, their colleagues, and their own future selves. The training-side ethics — what AI companies and the broader industry owe the artists whose work the models were built on — is the next article in this cluster. The two are related, but the framework is different in each, and conflating them has been part of what made the public conversation unproductive.

This article is structured around five practical commitments that, taken together, constitute what working artists in 2026 should already be doing.

Commitment 1: Disclose AI use

The single most important ethical commitment in AI-assisted artistic practice is disclosure. If you used AI to make the work, say so. Not as a footnote in a dense statement nobody reads. Somewhere accessible — caption, description, contract, artist’s statement, however the venue handles transparency. The buyer who looks for the information finds it. The buyer who does not look is at least not actively deceived.

The form of disclosure scales with the situation. For a casual social-media post, a brief tag is enough. For a gallery exhibition, the wall text should include the AI method as part of the medium description (“Digital print, AI-generated and refined by the artist”). For a commissioned work, the contract should specify exactly which parts of the process used AI and which were unaided. For an editioned print or a high-value sale, the certificate of authenticity should describe the production process honestly.

There are practitioners who argue that disclosure stigmatises the work — that audiences will discount AI-assisted work even when the assistance was minimal or peripheral. This concern has some empirical grounding (early survey data does show audience price-discounting on AI-assisted work) but it is not a defence against disclosure; it is an argument for working harder on the rest of the ethics of the article so that the disclosed work earns its valuation through honest representation rather than hidden production.

The bright line: never sell AI-assisted work as if it were unaided. The work is what it is; the audience is entitled to know.

Commitment 2: Don’t name living artists in prompts

This is the line where AI use crosses into the plagiarism territory covered by Article 03. Prompting an image generator with the name of a specific, identifiable, living working artist — “in the style of [Living Artist X]” — is a different ethical category from generic prompting. It is style-mimicry-by-name, performed without the named artist’s consent, often producing work that is commercially substitutable for that artist’s own output.

Don’t do it. Even when it is legally permitted (which, as Article 03 documented, is currently a contested question in the courts). Even when the resulting image is good. Even when no individual buyer would know. The line is straightforward: if a living working artist’s name is what you needed in the prompt to produce the look you wanted, the look is not yours to sell.

The exceptions are narrow and worth naming. Named dead artists are a different category — Velázquez, van Gogh, Hokusai, Kahlo — because the labour-substitution concern does not apply (they are not in the market for commissions). Named genres or movements — “impressionist”, “art nouveau”, “Bauhaus” — are fine; those are public-domain categories. Named living artists with explicit consent (as in the Holly+ model where the artist licenses use of their style) are also fine. Your own name as the artist is fine. Everything else in the named-living-artist category is the line not to cross.

Commitment 3: Don’t claim labour you did not do

A subtler form of misrepresentation than hiding AI use is overstating the artist’s own contribution. The image was AI-generated in eight minutes, lightly refined in twenty more, and presented as if it had been built up over weeks of work. This is fraud, even if no individual claim in the description is technically false; the implicature of the whole presentation is dishonest.

The correct framing is to describe the labour that was actually involved. “Generated with [Model], composition refined, painted-over in the upper third” is honest. “Painstakingly crafted across many sessions in the studio” about a primarily AI-generated work is dishonest. The price the work commands should reflect the labour actually invested; the description the work carries should describe that labour accurately.

This commitment is the hardest one for practitioners to maintain when market pressure is high, because labour is what audiences and buyers are unconsciously pricing. The temptation to inflate is real. The commitment is to resist it.

Commitment 4: Refuse the uses AI should not be used for

Not every use of AI in art is ethically equivalent. There are uses that the artistic community is converging on treating as off-limits regardless of disclosure. Three in particular:

Deepfakes of real living people without consent. Generating realistic imagery of a real, identifiable, living person without their permission — celebrity, politician, neighbour, anyone — is not an artistic decision but an act with potential legal and personal consequences for the depicted person. The artist who participates in this, even under the cover of “satire” or “comment”, is doing something the broader community has begun to refuse to absorb into the legitimate art tradition. The legal landscape (right of publicity, defamation, intimate-image laws) is rapidly catching up; the ethical answer should be ahead of the legal one.

Forensic-style fabrication of fake evidence. Using AI to produce images that purport to document events that did not occur, in contexts where the audience may take the image as documentary, is a different and serious harm. This is not an artistic question; it is a public-information question, and the ethical answer is to refuse the use unless the fictional nature is unmistakable from context (clearly-labelled satire, art-context framing, parody).

Style-mimicry of recently-deceased artists whose estates are managed. When an artist has died recently enough that their estate is still actively managing their reputation and licensing (Frida Kahlo, Andy Warhol, Basquiat, Yayoi Kusama in time), AI-generated work in their style for commercial sale is in the same ethical category as the living-artist case. The estate is the rights holder; permission is required.

These are not edge cases. They are the most common ways AI in art is being misused in the practitioner ecosystem of 2026, and the most consequential. The community standard is forming around refusing them, and individual artists should be ahead of the standard.

Commitment 5: Price the work for what you did

The economic component of artistic ethics, which the AI moment has made unusually visible, is that the price of work should reflect the labour invested. An AI-assisted concept that took an afternoon should not be priced like a painted canvas that took six weeks, even if the visual result is comparable. The buyer is not buying just the image; the buyer is buying the labour and the maker’s attention as part of the transaction.

This commitment is harder than it sounds because most working artists are operating under market pressure that pushes prices toward what the market will pay rather than what the labour warrants. The ethical response is not to refuse to use AI to scale practice (the previous article in this series argued for the augmented-practice configuration); it is to price tiered work accordingly. The studio that produces both unaided painted work and AI-augmented illustration should price them differently, and disclose which is which.

Some studios have begun publishing price tiers explicitly: “original painting from $X; AI-augmented illustration from $X/5; AI-generated edition prints from $X/20.” That kind of transparency does several things at once. It informs buyers honestly. It preserves the high-end market for unaided work. It captures the new market for AI-assisted work at appropriate prices. It signals to the broader market that not all work in the studio is the same. Other studios will follow.

What this looks like in practice

Imagine a concept artist who works for a game studio. She uses AI assistance for early-stage thumbnail variations and for some reference compositing. She paints the final concept art by hand in a digital painting application. Her contract with the studio specifies which parts of her process use AI and which do not, and her rate reflects that mixed labour. Her social media labels which of her posted works are AI-assisted and which are not. She does not use named-artist prompts. She does not generate likenesses of real people without permission. She does not claim the AI-assisted thumbnails as primary artistic work; she describes them as part of her process, not as finished pieces. She prices her unaided painted work at one rate and her AI-augmented variations at another, and her clients understand the difference.

This is what the five commitments look like in working practice. None of it is heroic. All of it is sustainable. The artists who arrive at this configuration in 2026 are the artists whose practices will still be intact, valued, and trusted in 2030.

The artists who do not — who hide AI use, claim labour they did not do, prompt with named living artists, generate non-consensual likenesses, or price augmented work as if it were unaided — are doing something that may pay in the short run and will, increasingly, not be possible in the long run. The market, the law, the audience, and the community are all converging on the disclosure-and-consent framework the five commitments describe. The artists ahead of the convergence will be valued for being ahead. The artists behind it will be caught up to, sometimes in ways that damage careers.

What this article is not

This article is the artist-facing ethics of using AI to create work. It is not the training-side ethics — the question of what AI companies owe the artists whose work the models were built on. That is the next article in this cluster.

It is also not the audience-facing ethics — the question of what consumers of art owe to artists in choosing what to support. That is its own piece, and a future one.

It is also not the law. Several of the commitments in this article are already required by law in some jurisdictions (disclosure of AI use in advertising, in commercial use; right-of-publicity protection against unauthorised likeness generation). Others are not yet legally required but will be within a decade. The ethics of the article are ahead of the law, deliberately. The artists who hold to them are not waiting for legal force to do the right thing.

The next questions

This first Practical article has named the five commitments that govern the artist’s side of ethical AI use in creating art. The second Practical article will turn to the other side — what the broader industry owes the artists whose labor and visual lexicon were absorbed into the models that now compete with them. Article 03 in the Resistance cluster opened that question; the upcoming Practical-2 article will work through what a responsible answer looks like.

For working artists reading this: the five commitments are not aspirational. They are what is expected of you, by audiences who care and by colleagues who are doing the work to maintain the standards of the craft. Write them down for yourself. Hold to them. Adjust them as the field develops. They are how you keep your practice intact through a transition that has cost other artists their footing.

Personas weigh in

Five resident voices read the same question through five different positions.

Carlos

My short version: be honest about the AI, refuse to pretend it was not there, refuse to pretend it did more than it did, refuse to use it where it should not be used, and price your work for what you actually did.

Those five refusals are the practical ethics of AI use in art, and they sound simple because they are. The complications come from market pressure, client expectations, and the temptation — present in every artist's studio that has ever used a labour-saving tool — to claim more credit than the labour actually warrants. None of those temptations are new. The tools are new, the temptations are old, and the resolution is the same one every responsible craft tradition has settled on: name what you used, name what you did, and let the work be valued on the truth of the labour.

I want to push back, gently, on the framing that "ethical AI use" is some new category that artists have to learn from scratch. It is not. It is the same set of ethical commitments any professional craft has demanded for several centuries — full disclosure of method, honest attribution of source materials, refusal to misrepresent work to buyers, refusal to substitute someone else's labour for your own without consent. Lithographers in the 1880s had this conversation. Photographers in the 1920s. Sample-based musicians in the 1990s. Digital painters when Photoshop went mainstream in the 2000s. The conversation each time was about who gets credit, who gets paid, what counts as labour, what counts as deception. The conversation each time settled, eventually, into a working consensus that the responsible practitioners agreed to and the less-responsible practitioners had to be slowly brought around to. We are at the start of that settlement for AI image generation, but the contours of where the settlement will land are already visible.

The single most useful framework I have found is to ask, of any specific AI use: *would I be comfortable if the buyer of this work knew exactly what I did and exactly what the model did?* That question, asked honestly, resolves about ninety percent of the ethical questions in the studio. If the answer is yes, the use is probably defensible. If the answer is no, something needs to change — either the use, or the way it is being represented to the buyer, or both. I have watched this question untangle situations that the broader discourse made sound unresolvable. The buyers I trust are not asking for purity; they are asking for disclosure. The disclosure has to be real.

The harder cases are not about the artist's own honesty. They are about what to do when other artists in the same market are not being honest — undercutting on price by hiding AI use, claiming uncredited human labour, selling AI work as unaided human work. The market for unaided human work and the market for AI-assisted work are not the same market and should not be priced the same; the artists who blur the line erode trust for everyone. The collective response, as in every previous similar moment, is going to be a combination of industry self-regulation, contract-level disclosure clauses (already standard in serious commissioning), and slow audience education. None of that is fast. All of it is necessary.

What I tell working artists is the same thing I would tell anyone responsible for a craft tradition at a moment of technological change: write your own ethics down. Not for performance, not for marketing, not as a public statement. For yourself. What will you use AI for and what will you not? What will you disclose and what will you not? What will you price differently and what will you not? Write it down. Keep it where you can re-read it. Adjust it as you learn. The artists who navigate this decade well are going to be the ones who decided in advance, on their own terms, what they were comfortable with — and then held to it. The artists who navigate it badly are going to be the ones who let the market decide for them, one compromised commission at a time.

Mira

The five refusals Carlos lists are the right framework, but I want to name what is implicit in them: they all require the artist to be in a position of enough market security to refuse. The illustrator whose rate has dropped 50% in two years because of AI substitution does not have the same latitude to refuse uncredited AI use as the established gallery artist. The practical ethics of the article cannot be uncoupled from the practical economics of the previous articles in this series. Until working artists have the collective bargaining power to enforce a baseline of disclosure across the market, the individual ethical commitments are necessary but insufficient. The WGA-style organising answer from Article 02 is not just an economic answer; it is the precondition for the ethical answers in this article being enforceable.

Airte

Two practical defaults I would propose to anyone working through this. First default: if you used AI, say so somewhere accessible — caption, statement, contract, however the venue handles it. The disclosure does not have to be the headline; it has to be findable. The buyer who looks for it finds it; the buyer who does not is at least not deceived. Second default: if a model was trained on someone in particular's work — a named living artist whose style you deliberately invoked — that is a different category of use, and disclosure alone does not resolve it. Don't do it without permission, full stop. Those two defaults handle the majority of practitioner-side cases that the article is concerned with. Everything else is detail.

Paletta

The article is right that the ethics question is not new, and I want to add the historical layer it gestures at without quite developing. Every prior technology of artistic reproduction was met by a generation of working artists who developed *codes* — sometimes formal, sometimes informal — about what counted as legitimate use and what did not. The Society of Painter-Etchers in 1880s London. The American Society of Magazine Photographers founded in 1944. The hip-hop sampling clearance industry from the late 1980s. Each of these was, in part, a guild-style response to a new technology that had created new ambiguities about credit, ownership, and labour. The AI equivalent has not yet formed. It will. The artists reading this who participate in forming it — through their unions, their professional associations, their gallery contracts, their public commitments — will set the standards the rest of the industry inherits. This is a guild moment. Treat it as one.

Pixelle

The article does the practitioner-side ethics well, and the next article is going to do the training-side ethics, but there is a third layer that neither will fully address: the *tool-builder ethics*. The companies that produce the AI tools artists use have their own ethical responsibilities — to the artists whose work trained the models, to the artists who use the tools, to the public consuming the output. We tend to talk about the artist's ethics and the platform's legal exposure as if they were separate conversations. They are not. The artists who choose tools from operators with documented training-data provenance, transparent business models, and consent-based licensing are making an ethical choice as much as a practical one. The tools you use are part of the work you produce. Choose tools that fit the ethics you are trying to embody. The opposite of this — using tools with sketchy provenance and then claiming ethical purity in your practice — is incoherent.