Plagiarism & Use of Large Language Models

Rajeshwari Hariharan | November 2025

I the last few years, the concept of authorship has dramatically changed. The catalyst for this change has not been a legislative intervention or statutory overhaul. It has been the sudden availability of extremely capable Large Language Models (“LLMs”) — GPT systems, Gemini, Claude, Llama etc that generate fluent, structured, well-reasoned text in seconds. Because this writing looks like human academic prose, it is hard to distinguish and much of the old vocabulary of plagiarism was not built for this moment. Thus, this technological intervention makes us question the very basis of “originality” and “creativity”. What is creativity today? There is a question mark on authorship – what is authorship today? Can such an AI-ghost written work be infringed?

Traditional plagiarism had a simple logic: A wrote X- B took X and presented it as their own. A being the author of work X was aggrieved on account of B presented it as his own or copied large parts of X. Today, we have a new species of text: work that may not be written by a human who is claiming authorship and also may not be copied not from any single author. This creates the zone of debate: synthetic plagiarism +authorship.

An author under the Copyright Act 1957 is the person creating that work. Thus, if a person creates a piece and with technological assistance, the language is polished, and one could argue that it is a positive technological intervention for creation of a better output. On the other hand, presenting a paper that has emerged from a command “write a paper on X topic” without much human intervention (in this era of genitive AI) is the subject of debate, since it is a work not created by the author at all.

What is “original” – this is not defined by the Copyright Act. However, under Designs Act 2000 – original means a work that has originated from the author. In other words, a work was not copied in part or totally from another or a third party or another source.

As we speak, right from academic thesis to blogs to articles are written with the active aid and assistance of machines and many times, as the language is so close to human prose writing that it is nearly impossible to detect. Thus, a fundamental question arises – what is originality and whether the work is an original work at all. In works made using AI assistance, this important debate is brought into sharp focus.

If AI is used as labour, it does not matter as there is no mandate in law that all labour going into a work should be human only. It matters whether the person submitting the work is “falsely claiming authorship” – pretty much equivalent to hiring a ghostwriter – also the ‘originality’ claim is probably incorrect, because some of the ideas in the work may have been AI generated.

We now have two separate axes of originality — (i) originality of idea and (ii) originality of language. A writer may appear “original” merely because the words are novel, while the underlying work is just recombination of pre-existing ideas. Since Copyright Act does not protect “ideas” per se, this new recombined work would qualify as a copyrightable work. If the contribution of AI is more than just labour, i.e a more intellectual contribution, then perhaps it would be most appropriate to indicate so in the article. But, surely, the work is not an entirely “original” work of the author.

ARE WE REDEFINING AUTHORSHIP ITSELF?

For centuries, authorship meant “expression of thought by a human.” But accepting the cultural shock that LLMs can not only express you’re your thought — but generate ideas you did not have will take sometime to digest.

The calculator analogy given by proponents of LLMs is useful but incomplete because with calculators we outsourced arithmetic; with LLMs we outsource thinking, searching, summarising, synthesising. Arithmetic is mechanistic. Thought is not.

SO WHAT IS THE STABLE RULE GOING FORWARD?

AI use is not wrong in law. Whether AI use without disclosure is wrong in law is debatable because the current Copyright Act 1957 has no provision to challenge authorship or originality. And it matters not in law as to how the author arrived at the work.

With the pace at which AI invasion is proceeding, eventually, using LLMs will become normalised — like spellcheckers once were — but the responsibility of idea custody will fall on the human. And “originality” may undergo a sea change.

COPYRIGHT INFRINGEMENT:

Many assume that “if AI wrote it, then nobody owns it” and therefore copying it cannot be infringement. That is legally incorrect.

Under most copyright systems, including India, the U.S. and EU — the law treats fixation and ownership as two separate layers. All said and done, the work is owned by the creator regardless of how he arrived at it.

This means:

the AI itself may not be an author under Copyright Act
the user who generated and obtained that text would own the copyright in that output Therefore, a third party who copies that text later (without permission) can still infringe copyright in that work generated with assistance of AI or by AI— even though the text originally came from an AI party or largely.

The misconception comes from conflating and confusing authorship with ownership. Authorship is about who created the expression. Ownership is about who holds the exclusive right to reproduce, distribute or communicate that expression. Terms of use of most modern LLM providers assign the output rights to the user. So, although the LLM model is not a legal author, the output can still become owned by the user who generated it.

Thus, if X uses an LLM to produce a paragraph, and Y later copies that paragraph into a paper or book without permission, Y is committing infringement — even if the paragraph came from non-human generation. This is because copyright protects against unauthorised copying of the expression, regardless of whether the underlying author was a writer, a photographer, or a prompt-designer directing an AI. LLM output is not automatically “public domain.”

Plagiarism in the AI era is not primarily about copying text but about misrepresenting authorship. Unfortunately, there is no specific provision to challenge “authorship” except by filing a rectification under the Copyright Act – but with what evidence? How does one prove that a work was created by an LLM or human. It is almost impossible to discern such nuanced difference.

The only sustainable culture and way forward is:

Mandatory disclosure of extent of AI’s role in the work;
Enabling challenge to authorship or expanding the definition of ‘originality’ to include works generated with AI
Recognizing infringement of “ideas” – that way humans will have to intervene in order to save their works from being labelled as “plagiarised work”

It is also the ethical equilibrium that we will need to consider whereby LLMs can co-exist with scholarship and copyrightable works— we should use LLM not as counterfeit brains — but as instruments that make human thinking more precise, not less human.

About Author

Rajeshwari Hariharan

Rajeshwari Hariharan is an accomplished advocate and arguing counsel before the Delhi High Court, and the founder of Rajeshwari & Associates, a boutique firm renowned for its expertise in intellectual property law. With a distinguished career spanning key patent, trademark, and trade secret disputes, she has played a pivotal role in shaping India’s IP jurisprudence, including her prominent involvement in the country’s first-ever compulsory licence case. A seasoned litigator, Rajeshwari has represented clients in complex matters before the Supreme Court of India, multiple High Courts, the Patent Office, and various tribunals. Her extensive experience includes handling high-stakes patent oppositions, technology-driven disputes, pharmaceutical IP matters, and cutting-edge issues involving confidential information and innovation-driven industries. Rajeshwari is also a prolific writer and speaker. She has authored numerous articles on intellectual property and related legal developments, and she frequently shares her insights at national and international forums. She is widely regarded for her deep patent expertise, strategic litigation skills, and effective courtroom advocacy, making her a sought after counsel in high-value and technically complex IP cases.