AI and Plagiarism in Your PhD: What the Rules Actually Require in 2026

Every research scholar in India is now using AI in some form, whether to polish a paragraph, understand a concept, or draft a first version of a section. The anxiety that comes with it is just as common: is this allowed? Will my university penalise me? Will a journal reject my paper because I used a tool? The honest answer is that there is no single "AI rule" that governs your PhD. There are two separate rulebooks, and you have to satisfy both.

The first is the plagiarism regulation your university is bound by. The second is the AI policy of the journal you submit to. They ask different things of you. This guide sets out what each actually requires in 2026, in plain terms, and where the genuine uncertainty still lies.

The rule that actually binds you in India

The binding instrument for Indian higher education is the UGC (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018. It does not mention AI, because it predates the current tools, but it governs the thing AI most affects: the originality of your text. It works on measured similarity, and it sets graded penalties.

These are the thesis and dissertation penalties for students. Faculty and researchers face a separate and heavier set of sanctions. Your own university ordinance may also set a stricter internal limit than the UGC floor, so always read it.

Importantly, the regulation excludes certain things from the similarity count, so a high raw number is not always what it seems. Quoted material reproduced with proper attribution or permission is excluded, as are your references and bibliography, table of contents, preface and acknowledgements, all generic terms, laws, standard symbols and standard equations, and common-knowledge or coincidental strings of up to fourteen consecutive words.

Why AI text is a plagiarism risk, not a loophole

Some scholars assume that because AI generates fresh text, it cannot be plagiarism. That is a dangerous misreading. A language model is trained on existing writing and often reproduces phrasing that is very close to published sources, which a similarity checker will flag. Worse, it can fabricate citations and results that look authentic but do not exist, and presenting invented references or data is a research-integrity violation far more serious than a similarity score.

The regulation holds you responsible for the originality and integrity of what you submit, regardless of how it was produced. "The AI wrote it" is not a defence. If anything it compounds the problem, because you have then submitted text you did not verify.

What journals now require

The second rulebook is the journal's. Major publishers and the Committee on Publication Ethics have converged on a clear position over the last two years, and most Scopus and Web of Science journals now state it in their author guidelines. Four points matter most.

A typical disclosure, following the format publishers such as Elsevier recommend, reads along these lines: "During the preparation of this work the author used [tool] in order to [reason]. After using this tool the author reviewed and edited the content as needed and takes full responsibility for the content." Always check the exact wording your target journal asks for, because it varies between publishers.

Is there a specific "UGC AI rule" for 2026?

This is where you should be careful, because much of what circulates online is not official. As of now, the binding UGC instrument on this subject remains the 2018 plagiarism regulation. A number of coaching and thesis-service websites describe a distinct "UGC AI guideline 2026" with specific AI limits and thresholds, but that could not be confirmed against an official UGC source. Treat those specific figures as unverified until you see them on the UGC's own website or in a Gazette notification.

What is genuinely happening is that individual universities are beginning to add their own AI-use and AI-detection norms on top of the UGC floor. That means the rule that actually binds you may be your own institution's ordinance, not a national one. Before you rely on any number, confirm it in your university's PhD ordinance and on the official UGC website.

Using AI safely: a line to hold

None of this means avoiding AI. It means using it as an assistant, not an author. As a working rule, the safe uses are the ones where you stay the thinker: brainstorming angles, understanding a difficult concept, improving the clarity of sentences you wrote, and checking your own grammar. The risky uses are the ones where you outsource judgment: generating findings, letting a tool invent citations, or submitting passages you have not verified against real sources and rewritten in your own voice.

Three habits keep you on the right side of both rulebooks. Verify every fact and citation a tool gives you against the actual source. Rewrite AI-assisted text in your own words and run your similarity check early, not at submission. And disclose substantive AI use wherever the journal asks for it. Do those three things and AI becomes what it should be: a way to work faster, not a way to lose your degree.

Conclusion

There is no single AI rulebook for your PhD, and waiting for one is the wrong strategy. Two rules already govern your work: your university's plagiarism regulation, anchored in the UGC 2018 framework, and the AI-disclosure policy of the journal you submit to. Satisfy both, keep your similarity genuinely low, verify everything a tool tells you, and disclose when asked. The scholars who get into trouble are rarely the ones who used AI. They are the ones who let it think for them.