Citation and honesty

The norms around using generative AI in school, at work, and in public are still being written. In 2026, what counts as legitimate use and what counts as cheating depends on which institution you ask, which country you ask in, and sometimes which professor or manager. Most rules are partial. Many are contradictory.

This section is not about following one specific rule. It is about the underlying principle — honesty about what work the model did and what work you did — that will hold across whatever specific policies you encounter.

The simple principle

A short version:

Be honest about which parts of your work the model helped with, and stay within the disclosure expectations of whoever you are turning the work in to.

That’s it. The complexity comes from the second clause — different contexts have different expectations, and they are evolving fast.

Schools and universities

The rules are sharpest, and most actively contested, in education.

General principle. Most institutions distinguish between using AI to learn (often allowed) and using AI to produce work that is then submitted as your own (typically called plagiarism, even when the AI’s output is not directly anyone else’s writing).

Practical rule of thumb. If a professor or institution has not explicitly authorised AI use for an assignment, assume it is not authorised. Ask before submitting AI-assisted work, even when you think the use was minor.

Specific cases worth knowing:

• Drafting an essay in your voice from your notes, with the AI cleaning grammar, is usually acceptable.
• Asking the AI to write a paragraph that you then submit unchanged is usually plagiarism, even when the AI didn’t copy anyone else.
• Asking the AI to brainstorm thesis ideas is usually acceptable, with disclosure.
• Asking the AI to summarise a paper you then cite as if you read is usually deeply problematic.
• Translating your work between English and Nepali is usually acceptable, with disclosure.

The pattern: the more the final intellectual content of the submitted work is the model’s and not yours, the more problematic the use is.

When in doubt, disclose. “I used ChatGPT to brainstorm structure and check grammar; the content and analysis are my own.” A teacher who minds your disclosure will tell you. A teacher who didn’t know is no longer surprised by the disclosure later.

Workplaces

Norms are more variable. Some companies prohibit any external AI use. Some encourage it actively. Most are somewhere in between, with policies that don’t exist or are ambiguous.

Practical advice:

1. Find your organisation’s policy. If it doesn’t exist, ask whoever sets policy. If they don’t know either, propose one.
2. Disclose AI use proactively to clients unless you’re sure they’re fine with it. “I used AI to draft this report; I’ve reviewed and edited it for accuracy” is usually well-received.
3. Be especially careful with client-facing or attributable work. Your client may not realise their report was AI-assisted; their willingness to pay full rates may depend on knowing.
4. Keep records of which work was AI-assisted. If a regulator asks two years from now, you want to be able to answer.

For Nepali professional contexts: the general norm in 2026 is that AI-assisted work is acceptable in most contexts with disclosure, but expectations vary. Government work, journalism, legal practice, and medicine have stricter disclosure norms; private business is more relaxed.

Journalism, publication, and public-facing writing

The strictest disclosure norms. In 2026, most reputable Nepali and international publications have explicit policies:

• AI-generated text typically must be labelled.
• AI-generated images often must be labelled, especially in news contexts.
• AI-generated quotes are usually banned outright (with the exception of using AI to translate or transcribe).
• AI-assisted research is generally acceptable but must be verified before publication.

The reason: a publication’s credibility depends on its accountability for the content. AI generation breaks the chain of accountability in ways readers care about. The discipline is to be upfront.

For Nepali journalism in particular, this matters because trust is already fragile. Publishing an AI-generated image of a politician without disclosure — even if the depiction is essentially accurate — damages the credibility of every subsequent piece of reporting from the same outlet.

Code, design, and creative work

Norms are still settling.

Code. Most open-source projects and many companies now accept AI-assisted code with disclosure (often as a commit message tag). For client work, disclose. For your own work, your decision.

Design and art. A genuinely contested space. AI-assisted art is acceptable in some contexts (concept design, commercial work) and rejected in others (fine-art contexts, juried competitions). The fast-changing norms make this one to watch.

Music. Similar. AI-generated music is acceptable in some commercial and demo contexts, contested in others.

The pattern across these: as the underlying technology improves, the question of how much human involvement makes a work appropriately yours becomes finer. There is no universal answer. The principle remains: disclose what you did, what the model did, and let the recipient decide whether they care.

The detection question

A common worry: “Can my professor / employer / publisher detect if I used AI?”

The honest answer in 2026: detection tools exist and are unreliable. They produce false positives (flagging human-written text as AI) and false negatives (missing actual AI use) at rates high enough that no serious institution should rely on them alone.

The practical implication: do not rely on not getting caught as a strategy. The detection tools are bad now, but they may get better. The reputational cost of being later discovered is asymmetric — your past work will be re-examined, and once trust is broken it is slow to rebuild.

The simpler approach: disclose, in advance, the help you received. This is also a more enjoyable way to work — without the anxiety of concealment.

A working personal policy

If you don’t know what to do, a reasonable default:

1. For any submitted, published, or client-facing work, include a brief note about AI use. “I used Claude to draft the structure; I wrote the analysis; I used DALL·E for the hero image.” Two lines, maximum.
2. For personal use (notes, drafts, exploration, learning), no disclosure needed. The AI is a tool like any other.
3. When the line is unclear, err on the side of disclosure. The cost is one sentence. The benefit is integrity.

This is not a complicated policy. It is, in 2026, probably the cleanest available one.

Check your understanding

Quick check

—

A student uses ChatGPT extensively to draft a university essay, with the model writing most of the analytical content. The submission instructions do not mention AI. The most defensible action is:

1. ASubmit the essay without disclosure — if the policy doesn't mention it, it must be allowed
2. BAsk the professor whether AI-assisted work is acceptable; if so, disclose the extent of AI involvement; if not, rewrite in your own voice
3. CSubmit without disclosure and hope detection tools miss it
4. DSubmit unchanged and tell the professor only if asked

Check answer→

What comes next

The final section. We’ve covered privacy and honesty; the last piece is workflow — how to integrate generative AI into your daily work over the long term without losing your own skill or judgement. We close the course there.