Copyright, Bias and Fair Use of AI Images and Text

By Ravindra Bagale, trainer · LinkedIn · Last updated: 19 October 2026

Generative AI course · Chapter 32 · Estimated reading time: 13 minutes

1. Suppose a bakery needs a logo and a job ad

Suppose our cousin Sneha opens Sneha's Oven, a small bakery in Nagpur. She asks us for two things this week: a logo and a job ad for a counter assistant.

For the logo, we type: "A logo like [a famous coffee chain's logo] but for a bakery, same green circle." The result is lovely, and also far too close to a famous trademark.

For the job ad, we type: "Write a job ad for a counter boy for a bakery." AI gives us "young, energetic boy, below 25, fair and smart-looking". We almost post it.

Both mistakes came from our prompts, and AI happily followed them. Let us understand the rules first, then fix both.

2. Who owns AI-made work?

This is a new and changing area of law. Here is what is clear so far, in simple words. This is general information, not legal advice; for a business decision, ask a lawyer.

United States

• The US Copyright Office said in its January 2025 report (Part 2, on copyrightability) that material generated purely by AI is not protected by copyright.
• Prompts alone are not enough to make you the author.
• Human contribution can be protected: your own text or artwork that you put in, your creative selection and arrangement of AI outputs, or your real changes to them.

India

• India's Copyright Act, 1957 says that for a computer-generated work, the author is "the person who causes the work to be created" (Section 2(d)(vi)). How this applies to modern AI tools is still being worked out.
• In August 2026, India's Registrar of Copyrights refused to record an AI system as the author of an artwork; it treated the human who caused the work to be created as the legal person to consider.
• Training on news content: in July 2026, the Delhi High Court refused an interim injunction that the news agency ANI had asked for against OpenAI. The court said that, at first view (prima facie), storing ANI's content to train ChatGPT fell within fair dealing under Section 52 of the Copyright Act. This was only an interim ruling; the main case continues.

The tool's own terms matter too. Each AI company says in its terms who may use the output and how. For example, some music tools keep ownership of songs made on their free plan (see the chapter on AI voice and music).

3. Fair dealing, fair use, and where they stop

• Fair dealing (India, Section 52) and fair use (US) let people use parts of copyrighted work without permission in some situations: research, private use, criticism, review, reporting.
• They are defences, decided case by case. They are not a blanket permission to copy.
• Court cases about AI training do not give you permission to copy someone's output: a photo, a cartoon character, a logo, a song or a page of text.

Safe habits:

1. Do not ask AI to copy a specific living artist, a brand's logo or a famous character for commercial use.
2. Do not upload someone else's photo or artwork and ask AI to "make it mine".
3. Describe a style in general words ("warm watercolour, soft light") instead of a person's name.
4. For text, write in your own words and quote with credit when you use someone else's lines.
5. If you would be embarrassed to show the original next to your version, do not publish it.

4. Why AI repeats bias

AI models learn from huge amounts of text and images written by people. So they also learn people's unfair patterns:

• Job words linked to one gender ("nurse" = woman, "engineer" = man).
• Ideas of beauty linked to fair skin.
• Age limits and "young" as a stand-in for "energetic".
• Some places, languages or groups shown in only one way.

When your prompt is vague or already biased, the model fills the gaps with these patterns. Bias in AI output is not only unfair; in a job ad, it also shrinks the pool of good applicants, and it can conflict with employment laws and company policies.

5. Worked example 1: a biased job ad, fixed

The first prompt gave us this ad:

Wanted: young, aggressive sales executive, below 25 years, male preferred.
He should have native English and his own bike. Rockstar performers get incentives.

Find the problem words. We wrote a small Python checker with a list of words that often exclude people:

# Flag words in a job ad that can put some applicants off or exclude them.
import re
FLAGS = {"he": "gendered pronoun", "his": "gendered pronoun", "young": "age", "below 25": "age",
         "rockstar": "vague, often off-putting", "aggressive": "can put applicants off",
         "native english": "excludes by origin", "fair": "appearance", "male": "gender", "female": "gender"}
ad = ("Wanted: young, aggressive sales executive, below 25 years, male preferred. "
      "He should have native English and his own bike. Rockstar performers get incentives.")
low = ad.lower()
for w, why in FLAGS.items():
    if re.search(r"\b" + re.escape(w) + r"\b", low):
        print(f"{w!r:18} -> {why}")

Real output (we ran this on 19 October 2026):

'he'               -> gendered pronoun
'his'              -> gendered pronoun
'young'            -> age
'below 25'         -> age
'rockstar'         -> vague, often off-putting
'aggressive'       -> can put applicants off
'native english'   -> excludes by origin
'male'             -> gender

Eight problems in two sentences. A word list is only a helper; it cannot understand meaning, so you still read the ad yourself.

The fixed prompt:

Write a job ad for a counter assistant at Sneha's Oven, a small bakery in Nagpur.
Describe the work and the skills only: serving customers, handling UPI and cash,
keeping the counter clean, basic English and Marathi or Hindi for talking to customers.
Use "you" and "they", not "he" or "she". No age, gender, looks or marital status.
If a bike is truly needed for deliveries, say "able to make local deliveries";
do not ask for one as a condition. Mention the working hours and that freshers are welcome.

Expected output (yours will be similar, not the same): a clear ad that lists the tasks, skills and hours, addresses "you", and says freshers are welcome. Read it once more for any word that describes a type of person rather than a skill.

6. Worked example 2: a licence check for a logo

Step 1: a clean prompt. No brand names, no famous characters:

A simple, friendly logo for "Sneha's Oven", a home-style bakery.
A rolling pin crossed with a wheat stalk inside a rounded shape, warm brown and cream,
flat vector style, readable at small sizes, no text inside the picture.

Add the bakery name yourself in a design app, so the letters are clean and correct.

Step 2: check it before you use it.

1. Reverse image search the logo with Google Lens. Does anything very similar come up?
2. Search India's trademark registry (IP India's public search) for the name "Sneha's Oven" and for similar names in the bakery and food classes.
3. Read the tool's terms: can you use the output commercially on your plan?
4. Keep records: the prompt, the date, the tool and every version, in one folder.
5. Make it yours: change the colours, adjust the shapes and set the name yourself. Your human changes also strengthen your claim to the final design.

Expected output (yours will be similar, not the same): a short checklist with a tick next to each step, and a folder with the prompt and 3 saved versions. If step 1 or 2 finds something close, change the design or the name before printing anything.

For a big investment (signboards, packaging, a franchise), ask a trademark lawyer to do a proper search. A proper search now costs far less trouble than repainting a shop later.

7. Worked example 3: credit lines that are honest

Credit lines tell people how a piece of work was made. They build trust, and some platforms, schools and companies require them.

Prompt to make your own:

I made this poster: my own photo of our cake, with an AI-generated background
from Gemini, and text I added in Canva. Write a one-line honest credit line for
the bottom of the poster, under 15 words.

Expected output (yours will be similar, not the same): "Photo by Sneha's Oven; background created with Gemini AI; design in Canva."

Creative Commons reminder: "CC BY" means you must credit the creator. "NC" in a licence means non-commercial only, so not for a shop poster. "ND" means no changes allowed. Always check the exact licence on the source page.

8. A responsible-use checklist for the whole course

You have now used AI for writing, data, images, video, voice and music. Before you publish anything made with AI, run this list:

1. Accuracy: did I check every fact, number and claim myself?
2. Privacy: did I keep private details out of the prompt?
3. Consent: do I have permission from every real person whose face or voice appears?
4. Copyright: did I avoid copying brands, characters, artists and other people's text?
5. Bias: does the output describe skills and facts, not types of people?
6. Licence: does my plan allow this use (especially commercial use)?
7. Label: have I said honestly where AI was used, where it matters?

9. Beginner mistakes and quick fixes

10. Try at home

Recap

Run the 7-point checklist on the next thing you publish with AI. Using AI well is not only about better prompts; it is about being the person others can trust with it.

Previous chapter: AI Detectors and Humanize AI Tools: What Works and What Is Honest

Next chapter: Vibe Coding for Beginners: Build a Website with AI (No Coding Needed)

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The bakery, people and job ad on this page are made up for learning. Legal points were checked on the US Copyright Office's 2025 copyrightability report, India's Copyright Act, 1957 on copyright.gov.in, and the Delhi High Court's 24 July 2026 judgment in ANI v. OpenAI on 19 October 2026; this is general information, not legal advice, and the law can change. The Python output was produced on 19 October 2026. AI output changes every time, so your output will be similar, not the same.