Where the Ethical Lines Are

Most AI ethics questions at work aren't dramatic. They're small and daily: Do I tell the client this draft started in ChatGPT? Is it fine to write a colleague's recommendation letter with AI? Can I use this generated image in an ad?

This lesson gives you a way to answer those quickly and consistently, so you're not deciding from scratch — or from guilt — every time.

---

1. The transparency test

Before using AI for a task, ask one question:

Would I be comfortable if the people affected knew exactly how AI was used here?

If yes, proceed. If you'd feel the need to hide it, that's the signal to change the approach or disclose.

This isn't a rule that you must always announce AI use. Nobody discloses spell-check. It's a test of whether hiding it would mislead someone who'd care.

2. When disclosure is expected

Context	Usually expected?	Why
Client deliverables	Yes, or per contract	They're paying for a process and may have data or IP rules
Academic submissions	Yes, per institution policy	Integrity rules
Journalism, research summaries	Yes	Readers rely on human verification
Realistic synthetic images, video, voices	Yes	Platforms and Indian IT Rules require labelling
Personal recommendations, condolences, letters in your name	Usually — or write it yourself	The value is personal effort and judgement
Internal first drafts you rewrite	Rarely	You're accountable for the final version
Grammar, formatting, translation help	Rarely	Minor assistance

When you do disclose, be specific and brief: "I used AI to draft the first version and summarise the source reports; I checked every figure and wrote the recommendations."

3. Informed consent

When AI is used on other people's work, data or likeness, they should know and agree:

• Clients — how AI is used on their project and their data. Put it in your proposal or terms.
• Colleagues — before recording and transcribing meetings with AI note-takers.
• Anyone whose voice, face or writing you'd clone or imitate — explicit, written permission.
• Audiences — when content they'd assume is human-made (a testimonial, a "personal" message) isn't.

4. The clear red lines

These aren't grey areas:

• Fake reviews and testimonials — deceptive advertising; illegal under consumer protection rules.
• Fabricated credentials, documents or evidence — fraud.
• Impersonating real people — voice or video deepfakes, fake messages — fraud, harassment, and under the 2026 IT Rules amendments, content platforms must remove.
• Misleading financial or health claims generated at scale.
• Non-consensual intimate or defamatory imagery of anyone.

5. Intellectual property — the unsettled part

The legal position is moving quickly and varies by country; verify the current position where you work.

• Ownership of outputs: many jurisdictions require human authorship for copyright. Purely generated output may not be protected; your meaningful creative contribution is.
• Training data disputes: authors, artists, news organisations and music labels have sued AI companies over training on their work. Outcomes so far are mixed.
• Output similarity: outputs can occasionally reproduce recognisable protected material, characters, logos or styles.

Safe practice:

1. Read the tool's terms for commercial use on your plan.
2. Don't prompt for named living artists', brands' or characters' look in commercial work.
3. Document your human contribution — drafts, edits, selection, composition.
4. For high-stakes commercial use (national ad campaigns, product packaging, book covers), consider licensed stock or original work, or get legal advice.

6. Misinformation and the environment

Misinformation: don't share AI-generated claims about health, politics, finance or current events without checking a reliable source. An AI summary of news can be confidently wrong.

Environment: training and running large models uses significant energy and water. For professional use this rarely changes the decision, but it's a reason to avoid wasteful habits — regenerating fifty images you don't need, or running an automation every minute when hourly would do.

7. The ethics decision tree

When unsure, work through:

1. Is it legal where I and the affected people are?
2. Could it harm anyone — financially, reputationally, emotionally — if it goes wrong?
3. Does it pass the transparency test?
4. Do the people affected know and agree where it involves their data, work or likeness?
5. Would I advise a friend to do the same?

Any "no" means stop, change the approach, or disclose.

---

⚠️ Common mistakes

• Treating disclosure as all-or-nothing. Match it to context.
• Using AI note-takers without telling meeting participants.
• "Testimonials" or reviews written by AI.
• Assuming you own everything the tool generates.
• Sharing AI summaries of news without checking.

---

What's next: the other side of transparency — how to tell whether content was AI-generated, and why detection tools can't be trusted on their own.