7. Ethics, rights and brand-safe use

One of the most important questions in generative AI is: who owns the output?

In most countries today, copyright law requires a human author. If a piece of art, text, or music is created entirely by an AI with no meaningful human creative input, courts and copyright offices in the United States, the European Union, and many other jurisdictions have found that it cannot be automatically protected by copyright. This means:

• Content you generate by simply typing a short prompt may NOT be fully protected as your intellectual property in many legal systems.

• The more creative control you exercise — writing a detailed prompt, selecting and editing outputs, combining multiple elements — the stronger your claim to the resulting work.

• The AI company that trained the model retains certain rights to the underlying technology and, in many cases, to the training data used.

For Ethiopian businesses and creators, this has three practical implications:

1. Do not assume you own all AI-generated content outright. Read the terms of service of the tool you are using — some grant you broad rights, others do not.

2. Adding your own creative work on top of AI output (designing the layout, writing original text, personalizing the image) strengthens your ownership claim.

3. Do not register AI-generated work as your own with agencies that require human authorship — this could amount to fraud.

Example: Abebe uses Midjourney to create a logo for his coffee export business. He cannot copyright the raw AI image. But if he heavily modifies it in a design tool and adds original elements, the final work may qualify for protection in some jurisdictions.

A second key area of rights involves what you feed into an AI tool — not just what comes out.

Input copyright rules

When you upload an image, audio clip, video, or text to an AI tool for editing or transformation, you need to own or be licensed to use that material. Uploading someone else's copyrighted photograph to an AI image editor and passing the result off as your own work is a copyright infringement — even if the AI has significantly changed the image.

Training data and ongoing lawsuits

Many generative AI models were trained on images, text, and music scraped from the internet without explicit permission from the original creators. This has led to lawsuits in the United States and Europe from artists, authors, and record labels against AI companies. As a user, you are not personally liable for how the model was trained — but you should be aware of the controversy, because some industries and clients may refuse to accept work created with certain AI tools.

Protected voices, likenesses, and characters

Never use an AI tool to:

• Clone a real person's voice without their written consent.

• Generate realistic images of a named real person without their consent — especially in commercial contexts.

• Reproduce copyrighted characters (Disney characters, branded mascots, etc.) in commercial content.

Safe inputs to use

• Your own original photographs, recordings, and text.

• Assets from royalty-free or Creative Commons licensed libraries (Unsplash, Pixabay, Pexels, Free Music Archive).

• Assets from stock libraries with a commercial licence (Shutterstock, Adobe Stock).

• Text and images you have explicitly licensed from their creator.

One of the most serious ethical risks in generative AI is the creation of misleading or false content — especially content that appears to be real.

What is a deepfake?

A deepfake is synthetic media — video, audio, or image — where AI makes it appear that a real person said, did, or appeared somewhere they did not. Modern AI tools can clone a voice from just 10–30 seconds of audio and create realistic talking-head video of someone saying words they never spoke.

Why deepfakes are harmful

• Political deepfakes spread disinformation, undermine elections, and can incite unrest. A fabricated video of an Ethiopian government official, religious leader, or public figure making a controversial statement could cause serious harm — and is illegal under Ethiopian defamation and misinformation laws.

• Commercial deepfakes — falsely showing a celebrity or competitor endorsing or disparaging a product — violate advertising standards and personal rights laws.

• Personal deepfakes — making intimate or embarrassing fake videos of private individuals — are a form of digital harassment, illegal in many jurisdictions.

Synthetic media disclosure requirements

Major platforms — Meta (Facebook, Instagram), YouTube, TikTok, X (formerly Twitter) — now require creators to label AI-generated content that could be mistaken for real footage of real events. Failing to disclose can result in:

• Content removal.

• Account suspension or permanent ban.

• Legal liability if the content causes harm.

Best practice: whenever you publish AI-generated images, audio, or video in a news-adjacent, informational, or commercial context, add a clear label such as: "This image/video was created using AI." (አማርኛ፦ «ይሄ ምስሉ/ቪዲዮ AI ተጠቅሞ ተፈጥሯዋል።»)

This is not just a platform rule — it is an ethical duty that protects your audience and builds long-term trust in your brand.

Beyond legal rules, every business using AI for content creation must think about brand safety — ensuring that AI-generated content reflects your brand's values, voice, and cultural context correctly.

What can go wrong?

AI models are trained on global data and may produce content that:

• Contains culturally insensitive imagery — for example, generating images of Ethiopian food, clothing, or ceremonies that look inaccurate or stereotyped.

• Uses language or humor that is offensive in an Ethiopian context, even if acceptable in Western markets.

• Includes unintended nudity, violence, or religious imagery that could harm your brand's reputation.

• Mixes up logos, color palettes, or visual elements with those of competitors — a particular risk when generating product mockups.

• Produces fictional claims about your product (e.g., AI-generated copy that says your product has certifications or features it does not actually have).

A brand-safety checklist for every AI output:

□ Does this text or image match our brand's visual identity (colors, fonts, style)?

□ Is the cultural depiction accurate and respectful for an Ethiopian audience?

□ Does the copy make any claims that are not factually true about our product or service?

□ Are there any religious, political, or sensitive symbols in the image that could cause offence?

□ Would this content be appropriate for all audience members we target, including different age groups and regions?

□ Does the output unintentionally resemble a competitor's brand?

□ Have we disclosed that this is AI-generated, where required?

Almaz's checklist example: She uses AI to generate product images for her hand-woven textile shop and marketing copy for TeleBirr payment promotions. Before posting, she checks: Do the AI-generated images show the traditional patterns accurately? Does the copy correctly describe her shop's Addis Ababa address and TeleBirr payment options? Has she avoided any cultural misstep? Only after this review does she publish.

To bring everything together, here is a simple five-step responsible AI content framework you can apply before publishing any AI-generated content for your business or clients:

Step 1 — Purpose check

Ask: Why am I creating this content? Is the purpose legitimate, honest, and respectful? If the content is meant to mislead, impersonate, or defame, stop immediately.

Step 2 — Consent and rights check

Ask: Do I have the right to use all inputs (images, audio, text, people's likenesses)? Do the AI tool's terms of service allow commercial use of the output? Are there any real people in the output who have not consented?

Step 3 — Accuracy check

Ask: Does all text in the output accurately describe my product, service, or claim? AI models can hallucinate — generate facts that are simply not true. Always verify factual claims before publishing.

Step 4 — Cultural and brand safety check

Ask: Is the output appropriate for an Ethiopian audience? Does it reflect my brand's voice and values? Could any element of it cause offence or embarrassment?

Step 5 — Disclosure check

Ask: Do I need to label this as AI-generated for the platform I am posting on? Is there any audience expectation that this is real human-created content that I am not meeting transparently?

If all five checks pass, you are ready to publish. If any check fails, fix the issue before posting — or create the content differently.

This framework applies whether you are generating an image for a small shop's Telegram post, producing an AI avatar video for a training course, or writing marketing copy for a brand campaign billed in birr.