Chapter 3 · Creativity, Work and Safety
Creative Rights, Imitation and Freelance Practice: Fair Work with AI
- Page 12 of 22
- 22 min read
In Copyright, Licences and Attribution: Using AI Output and Source Material you learned to separate copyright from plagiarism, to treat owning a copy and having permission as different things, and to check a licence before reusing an asset. This page moves from those basics to the situations a creator or freelancer meets every week: Who can claim a piece of work when AI helped make it? Where does inspiration end and imitation begin? What if a client asks for "the style of" a living artist? What can go wrong with AI-written code and text in paid work?
Several of these questions are open, and reasonable people disagree. The page does not give a verdict. It gives you the considerations, a few habits that protect you and the people whose work you touch, and a checklist you can reuse.
What you will learn
- Why the same AI-assisted work can be treated differently in different countries, and what to promise a client in the meantime.
- How to record your own creative contribution, and how to tell inspiration from close imitation.
- How to think through a request to imitate a living creator's style, and the risks of synthetic likenesses and of AI-generated code or text in paid work.
- How open-source licences differ from public-domain material, and the main arguments about creator pay, innovation and access.
- A copyright and attribution checklist for freelancers, applied to a client's unclear asset licence.
One running scenario
We will return to the same invented example through the page.
Invented scenario:
Santiago is a freelance designer in Bogotá. A small food company, Sol de Valle, hires him to make
a mascot, a shop poster and a short product text. Santiago uses an image tool for early mascot ideas
and a chat assistant for first drafts of the text. The client's owner, Ms Ferreira, writes: "Make
it look like the work of a well-known living illustrator we like. Also, we own whatever you make,
right? We don't want a competitor copying it." She also sends a folder of photos and a font,
saying "our old agency gave us these, they are fine to use".Almost every section below touches this brief. Note that Santiago is not doing anything unusual. The questions are about doing a normal job carefully.
Why copyrightability of AI outputs varies by country
Copyright is set by national law, and each country decides for itself who can be an author, how much originality is needed, and what counts as protected expression. Rules made for cameras, printing presses and word processors never had to answer a new question: if a system produced most of the expressive detail, whose work is it? Countries can answer that question differently, and some have not yet answered it fully. So the same image or text can be protected in one country and not in another, or protected only for the parts a person shaped.
This tutorial can show you one country's official view, dated and hedged, as an example of how such a view is built. It does not tell you what your country does.
Legal note (as of 2026-10-12; not legal advice): The United States Copyright Office published a report on copyrightability on 2025-01-29. It says copyright protects original expression in a work created by a human author, even if the work includes AI-generated material. It concludes that purely AI-generated material is not protected, and that protection is not available where human control over the expressive elements is insufficient. It states that, based on the functioning of currently available technology, prompts alone do not provide sufficient control. It adds that a human author is entitled to copyright in their own perceptible contributions, in the creative selection, coordination or arrangement of material, and in creative modifications, that the sufficiency of human contribution is assessed case by case, and that using AI to assist human creativity, rather than stand in for it, does not by itself affect protection. The Office says it will keep monitoring the area. This is one country's official report. It is not the law of any other country, and it can change. Check the current official text for your country and sector.
Two practical lessons follow, whichever country you work in.
- Do not promise what the law may not give. Ms Ferreira asks, "we own whatever you make, right?" and wants to stop competitors copying the mascot. If the AI-generated parts may not be protected where she sells, a competitor may be free to copy them. Santiago can honestly say: "I will transfer all rights I hold to you. Whether the AI-made parts are protected depends on the country, and I cannot guarantee it."
- Where the answer is unclear, ask. Name the country where the client will sell and the country where you work, then ask an adviser qualified there, or read your national copyright office's guidance. "Unknown" is a legitimate answer to write in a record.
Documenting your human creative contribution
When the amount of human contribution can matter, a record helps you in three ways: you can describe truthfully what you did, you can show a client the real structure of the work, and you can answer questions later. A record does not guarantee that any court or office will accept your contribution as enough. It is evidence of what happened, not a legal shield.
Keep it light. A few dated lines per project are enough, and you can write them as you go.
Template:
PROJECT: Sol de Valle mascot DATE STARTED: 2026-10-05
MY OWN WORK Hand sketches of the mascot's pose and proportions (files 01-04).
Colour palette chosen from the client's logo.
AI TOOL USED Image tool, for 12 rough variations from my text description.
Tool name, version and its terms as of 2026-10-05: [fill in].
WHAT I DID WITH IT Chose 2 variations, redrew the face and hands by hand, merged them,
changed all colours, added the lettering I drew myself.
WHAT STAYED AS IS Background texture from variation 7 (generated, not edited).
WHAT I TOLD THE CLIENT 2026-10-06, by email: AI was used for early ideas and the texture.Notice the line "what stayed as is". Being honest about the parts you did not change is what makes the rest of the record believable. It also tells you which parts a client should not assume are exclusively theirs.
Inspiration and close imitation
Every creator learns from others. Writers imitate sentence rhythms; illustrators study line and colour; musicians copy a groove while learning. Nobody owns a general mood, a genre or a technique. At the other end, a near-copy of one specific work is plainly taking. Most real cases sit between, so it helps to think in steps rather than as a switch.
| What the new work shares with the earlier one | Example | How to think about it |
|---|---|---|
| A general mood, genre or technique | "Warm, hand-painted, children's-book feel" | Normal influence. Nobody owns a genre. |
| A recognisable mix of recurring features, not tied to one work | A particular character design, colour habit and line quality together | A grey area. Ethical and sometimes legal views differ. Slow down. |
| Strong resemblance to one specific work or character | Same pose, composition and distinctive details | Close imitation. Seek permission or change it substantially. |
| A reproduction or tracing, even with small changes | An edited copy of the original file | Copying. Treat as needing permission. |
Three questions help you place a piece. Would a reasonable viewer think this is, or comes from, the other creator? Did I give the tool the other person's work as a reference, and if so, did I have permission? If the original creator saw this beside their work, would I be comfortable explaining my process to them? Legal tests for similarity can differ by country and case, so these questions are ethical checks. They do not tell you whether a court would find infringement.
Requests to imitate a living creator's style
Back to the brief: "Make it look like the work of a well-known living illustrator." Some people say a style is not property, so asking for it is ordinary. Others say a style is a creator's livelihood, and a tool trained on their work that can produce lookalikes in seconds takes that livelihood. Both groups include thoughtful people, and whatever rules exist on style depend on the country, so check them. Here are the considerations side by side.
| Consideration | Pushes towards "this is acceptable" | Pushes towards "hold back or ask" |
|---|---|---|
| What is being borrowed | A general look shared by many artists; ideas and techniques have always spread. | A distinctive, recognisable manner that took years to build and identifies one person. |
| Effect on the creator | Clients who want the original artist's price may not become your clients anyway; imitation can also raise their profile. | The client who would have commissioned the artist now commissions a copy; the creator gets no payment or credit. |
| How the request is made | Describing qualities (warm palette, loose line) without a name. | Naming the living person and asking for work that could pass as theirs. |
| Commercial context | A private mood board or early idea sketch has a small footprint. | A public logo, product or advert that sells on the association has a large one. |
| Consent and credit | If the creator agrees, or is hired, the question largely disappears. | No contact, no payment and no mention, while the tool may have learned from their work. |
There is no single right answer in this table. The decision is yours, and your reasons matter more than the label. Applying the course pattern to Santiago gives a short, defensible set of options.
Worksheet, filled in:
1. SITUATION Client asks for a mascot "like" a named living illustrator.
2. WHO IS AFFECTED Client, Santiago; the illustrator (no say); customers who may think it is theirs.
3. BENEFITS/HARMS Benefit: client gets the look they like. Harm: illustrator may lose work or
credit; Santiago's name is tied to a possible dispute.
4. OPTIONS A. Do it as asked, naming the person in the tool.
B. Ask the client for qualities (palette, mood, shapes) and design an
original look without naming the person or uploading their work.
C. Offer to commission or license work from the illustrator.
D. Decline this part of the brief.
5. EVIDENCE Known: client wants the feel, not a copy (to confirm). Unknown: whether the
illustrator has a stated policy on imitation; the rules in the client's country.
6. DECISION B, with C offered as the "exact look" route. Stop if the result is
recognisably that person's characters.
7. ACCOUNTABLE Santiago for what he delivers; the client for what they request and publish.
8. REVIEW DATE At the first draft: compare with the illustrator's public work.You might choose A, C or D instead. A good alternative decision still names the people affected and the condition that would make you change your mind.
Privacy and publicity risks in synthetic likenesses
A synthetic likeness is an image, video or voice that imitates a real person's face, body or voice. Two kinds of risk appear. The privacy risk: using someone's photo or recording as input to a tool shares their personal data and may produce content they never agreed to (see Privacy, Consent and Data Use: What You Share with an AI Tool). The publicity risk: a likeness of a recognisable person can suggest that the person endorses a product or says something they never said. Rules about using a person's name, image or voice commercially may apply where you work or where the work is shown. This tutorial does not describe them, so ask for local advice before using a recognisable person's likeness commercially.
In Santiago's job: Ms Ferreira suggests a spokesperson "who looks like a famous local football player". Santiago has several choices: use an invented character instead; hire a real actor who consents in writing to a defined use; or ask the player's representatives. Not asking is the weakest option. Likewise, using a staff member's photo as input needs that person's written agreement about what is made and where it appears. Deepfakes, Provenance and Disclosure: Telling Real from Synthetic explains how to label synthetic media so viewers are not misled. A label does not replace consent from the person shown.
Risks of AI-generated code or text in commercial work
Paid work raises the stakes, because a client relies on what you hand over. The risks below can be reduced, but not removed, by care.
| Risk | Why it matters | What you can do |
|---|---|---|
| Unclear protection and exclusivity | If AI-generated parts may not be protected, "exclusive" or "you own everything" may overpromise. | Use careful contract wording; say what you transfer; record your own contribution. |
| Text or code that resembles existing material | A model can reproduce phrases or code from material it saw. You may not see the source. | Review and rewrite; check distinctive passages with a search; for code, check licence terms of any matching project. |
| Licence obligations in code | Some code is shared under a licence with conditions. Mixing it into a client's product can create duties neither side expected. | Ask whether the client has a code-origin policy; keep a list of dependencies and licences. |
| Errors that look professional | Wrong claims in a brochure or bugs in code reach customers under your client's name. | Verify as in Prompt Engineering for AI: Beginner; test code before delivery. |
| Client confidentiality | Client files pasted into an outside tool leave your control. | Check the client's rules and the tool's terms; share the minimum or none. |
| Contract and tool terms | The client's contract or the tool's terms may limit AI use or commercial use of outputs. | Read both before you start, and note the date. |
Open-source licences and public-domain material
These two terms are often mixed up. Material under an open-source licence (or a similar licence for shared creative work) is still owned by someone. The owner has granted permission on stated conditions, such as giving credit, sharing changes under the same terms, or not using it for sale. Breaking the conditions can mean you have no permission. Public-domain material is not protected by copyright, either because protection has ended, was never available, or the owner has given up their rights. Whether a work counts as public domain can depend on the country, so a work may be free to use in one place and protected in another.
| Label you see | What it usually means | What to do |
|---|---|---|
| "Open source" or a shared-work licence | Owned, with permission on conditions. | Read the licence text; follow credit and sharing conditions. |
| "Public domain" or "no known restrictions" | Not protected, or the claim is unverified. | Check who says so and for which country; keep the source. |
| "Free to download" or "royalty-free" | Says nothing certain about permission. | Find the licence text; "free" and "licensed" are different. |
| No licence stated | Unresolved. Do not assume permission. | Ask the owner or choose another asset. |
Creator compensation, innovation and access
Behind many of these questions is a larger disagreement: how should the benefits and costs of AI trained on human work be shared? As Copyright, Licences and Attribution: Using AI Output and Source Material explained, courts in different places have not converged, and a settlement is not a ruling. The table below sets out strong arguments in each position, as commonly made by people in different roles and regions. The voices are invented to show perspective; the arguments are not facts about any named person or organisation.
| Concern | Strong argument from one side | Strong argument from another |
|---|---|---|
| Creator compensation (a voice actor in Mumbai) | Work that taught the system was made with time and skill. If it creates value, creators should have a say and a share, and some may lose income to cheap lookalikes. | (A start-up developer in Lagos) Requiring payment for every input could favour a few large firms that can afford it and shut out small ones. |
| Innovation (a researcher in Seoul) | Learning patterns from large amounts of text and images drives useful tools in health, translation and accessibility. Heavy restrictions could slow that work. | (An illustrator in Berlin) Innovation that depends on unpaid creative work is not neutral; progress can still respect consent and fair terms. |
| Access to knowledge (a librarian in São Paulo) | Open access to information supports education and research, especially where books and tools are costly. | (A publisher in Nairobi) Local authors and small publishers are fragile; weak protection can end the work that fills those libraries. |
| Practical control (a student in Dhaka) | Opt-out systems let rights holders say no, if they are easy to use and honoured. | (A photographer in Osaka) Opt-out puts the burden on creators who may not know it exists; permission first would be fairer. |
Each person above is partly right, and nobody here is asking for something unreasonable. These questions are being argued in courts, parliaments and industry groups, so do not treat any single answer as settled. For your own work, you can still make decisions: pay or credit people you borrow from, pick licensed sources where they exist, and be honest with clients about what you used.
A copyright and attribution checklist for freelancers
Use this at the start of a project and again before delivery. It takes about ten minutes and fits on one page.
Checklist:
BEFORE STARTING
[ ] Client's contract and rules on AI use read. Tool terms read (date: ______).
[ ] Country where the client will use the work: ______. Country where I work: ______.
[ ] Every client-supplied asset (photo, font, text, code, logo) has a licence I can read.
[ ] Questions about licences sent to the client in writing.
WHILE WORKING
[ ] I note which parts are my own work, which are AI-assisted, which are third-party.
[ ] I do not name a living creator in a prompt or upload their work without permission.
[ ] I do not use a real person's face or voice without their written agreement.
[ ] Client confidential material stays out of outside tools unless allowed.
BEFORE DELIVERING
[ ] Distinctive text, code and images checked for resemblance to existing work.
[ ] Facts and code verified or tested.
[ ] Credits and licence notices added where required.
[ ] A short note to the client: what AI was used for, and what I can and cannot promise.
[ ] The record saved with a date.Where this shows up at work
Agencies, in-house teams and sole traders all face the same moments: a client's brief asks for a look, an old agency hands over files, a deadline pushes you to paste text into a tool. A freelancer who has a written checklist, a short contribution log and a clear sentence about what they can promise usually avoids disputes and answers them calmly if they arise. Larger clients may add their own rules about AI use; in that case their rules come first. Work, Clients and Jobs: Using AI Responsibly Where You Are Paid to Decide returns to employer rules and disclosure.
Common mistakes
- Promising exclusive ownership. It may not be available for AI-generated parts. Fix: describe what you transfer, and say what you cannot guarantee.
- Treating "free to download" as permission. It tells you the price, not the licence. Fix: find the licence text, or ask.
- Naming a living creator in a prompt "just for inspiration". It can produce close imitation without you noticing. Fix: describe qualities and check the result against their public work.
- Using a client's staff or customer photos as input without agreement. People did not consent to this use. Fix: get written agreement about what is made and where it appears.
- Keeping no record. When a question comes up months later, memory fails. Fix: keep the short log and the client's emails.
Try it yourself
- Easy. Label each as OS (owned, with permission on conditions), PD (public domain claim) or U (unresolved): (a) a code library with a licence file saying you may use it if you keep a credit line; (b) a photo on a website with "free download" and no licence text; (c) a poem with a note saying its author died long ago and the poem is out of copyright in the site's country. For (c), say what you would still check.
- Medium. Aya, a designer in Osaka, is asked to make a poster "just like" a named living photographer for a local festival. Write two options, one that you would accept and one that you would decline, and give one reason for each. Then write one sentence Aya could send the client.
- Hard. Decide how to handle a client's unclear asset licence, using the scenario below. Use steps 2 to 8 of the eight-step pattern and write a short message to the client.
Invented scenario:
Meenal is a freelance web designer in Pune. Her client Tomás runs a tour company in Lisbon and
wants a new homepage. He sends a folder with 30 photos and a font file, and writes: "Our old agency
made these. We have a licence, it's fine." The folder has one text file, "licence.txt":
"Images licensed for editorial and internal use by the client. Not for resale. See agency."
There is no licence file for the font. Six photos show recognisable guests at a tour. Tomás wants
the homepage live in five days and also wants to use the photos in paid social-media adverts.
Meenal has used an AI tool to produce an extra banner image from her own sketch.Answers
Easy. (a) OS: owned, permission with a condition (credit). (b) U: "free download" with no licence text is unresolved. (c) PD claim: the poem may be out of copyright in the site's country, but you would still check the claim's source, whether your own country treats it the same way, and whether a modern translation or edition adds new rights. Pass: you separate permission-on-conditions from public domain and from unresolved, and you note that country and source matter. Weak: you treat "free" or "old" as automatic permission.
Medium. A strong answer: accept an option that describes qualities (mood, light, framing) and designs an original poster; decline an option that asks the tool to produce work that could be mistaken for the photographer's images, or that uploads their photos without permission. Reasons: the first uses general influence; the second takes a distinctive manner and risks a near-copy without payment or credit. Aya's sentence could be: "I can create a poster with this mood, but I would rather not copy a living photographer's work; would you like me to propose original options or contact them about a licence?" Pass: two distinct options, a reason for each, a polite client sentence, no unsupported legal claim. Weak: a flat verdict ("style is free" or "style is theft") without options or reasons.
Hard. A strong model answer. (2) Affected: Meenal and Tomás; the agency (rights holder, no say yet); the guests in six photos; website visitors. (3) Benefit: a homepage on time. Harms: use outside the licence terms, people shown without agreement, a font used without a licence, an AI banner whose protection is uncertain. (4) Options: A. Use everything as sent. B. Use only assets she can clear, ask Tomás to get a written licence for the rest, and replace the unclear ones with licensed or own material. C. Ask the agency for a licence for web and paid adverts. D. Decline the project. (5) Known: the text file says "editorial and internal use", which does not clearly cover a commercial homepage or paid adverts. Unknown: the font licence, whether guests consented, who the agency is. (6) Decision: B, with C requested. Conditions: no photo with a recognisable guest is published without written agreement; paid adverts wait until the licence is clear; she keeps the licence text with a date. Stop if the agency refuses. (7) Accountable: Meenal for what she builds; Tomás for the assets he supplies and what he approves. (8) Review: at day three, when the agency replies. Message: "Thank you for the files. The licence text allows editorial and internal use. Before I publish, I need a written licence covering the website and paid adverts, a licence for the font, and confirmation for the six photos that show guests. Meanwhile I will build the page with assets I can clear. I also used an AI tool for one banner, from my own sketch; I can send my notes." Pass: asks for evidence rather than assuming, separates the web use from the paid advert use, deals with the font and the guest photos, offers a workable path to the deadline, records AI use honestly, and states what Meenal cannot promise. Weak: "use it all, the client said it is fine", or refusing the whole project without a conversation.
Look back at the eight-step pattern from Everyday AI Ethics: Who Benefits and Who Could Be Harmed: the worksheet is the same, applied to a rights question.
Summary
- Copyright is national. Whether AI-assisted work is protected, and how much human contribution is enough, can differ by country. One country's official view is an example, not a rule for you.
- Do not promise exclusivity you may not be able to give. Record your own contribution, and be honest about the parts you did not change.
- Inspiration is normal; close imitation is not. Requests to imitate a living creator's style have fair considerations on several sides, and you can usually find an option that is safer for everyone.
- Synthetic likenesses, code and text in client work bring privacy, publicity, licence and accuracy risks. Open-source is not public domain, and "free" is not "licensed".
- Arguments about creator pay, innovation and access are real and unsettled. Use the checklist on every project.
Next: Learning, Integrity and Disclosure: Using AI in Study and Assessment turns to students and teachers: when AI helps learning, when it replaces it, and what to disclose.