A commercial detection tool can now label your writing as AI generated before a human reader has assessed it. Such a finding can then trigger That finding can trigger disciplinary proceedings at a university. It can lead an editor to reject your work. It can raise questions about the authenticity of professional writing you produced yourself. Indian law has not yet decided what evidentiary weight such findings should carry.
The debate around AI detection has focused almost entirely on whether the technology works. That is the wrong question. The more important question is whether a person should suffer any consequences based on the output of a probabilistic software, more so when no legal framework exists to gauge the reliability of such output, or to even challenge it. India has begun using these tools before answering either question.
The regulations stop at plagiarism
The University Grants Commission's (UGC) published it’s anti-plagiarism regulation1 in 2018. It sets similarity thresholds for academic submissions, and prescribes consequences for plagiarism. While large language models (LLMs) were not part of mainstream academic use at the time, the regulation remains silent on their usage even today.
Indian universities, however, have decided to forge their own path. Several IITs, NITs, and central universities updated their academic integrity policies over the past few years to require AI detection checks in addition to conventional plagiarism screening. Through Information and Library Network’s (INFLIBNET’s)2 ShodhShuddhi programme3, UGC-governed institutions get access to DrillBit Extreme for plagiarism and AI detection in relation to doctoral theses, research articles, thesis chapters, and synopses. The programme expressly excludes undergraduate and postgraduate assignments from its scope. Many universities have also independently incorporated Turnitin, GPTZero, or Originality.ai as part of their academic integrity processes.
None of this rests on any amendment to the UGC regulations, which define plagiarism consequences by similarity threshold and say nothing about AI detection scores because the tools did not exist when the framework was drafted. A similarity score identifies textual overlap with existing sources. An AI detection score attempts to predict authorship. These are different issues with different evidentiary requirements, and a document can clear every plagiarism threshold in the UGC Regulations while still being flagged as AI generated.
Many institutions now distinguish between disclosed and undisclosed AI assistance. This distinction reflects a growing position in academia and industry, that use of AI does not necessarily amount to misconduct. The legal uncertainty instead stems from how institutions establish that AI was used at all.
While Universities provide the clearest example, the same issue is surfacing wherever authorship matters. Publishers, employers, and professional bodies are all confronting questions about AI generated content, but none of them has settled what legal significance AI detection score should carry. Until this question is answered, institutions will continue treating probabilistic conclusions as findings of fact.
AI detection tools have attracted sustained criticism from researchers, and educational bodies. UNESCO has cautioned4 educational institutions against relying solely on AI detection systems because of persistent concerns about reliability and fairness. Research published in Patterns, a peer-reviewed scientific journal, found that leading AI detectors disproportionately flag writings by non-native English speakers as AI generated5. This attains particular significance in India, where English is the primary language of higher education for millions of students, who did not grow up speaking it.
Indian writing faces higher false positives
AI detection tools do not determine who wrote a document. They estimate whether a document resembles text commonly produced by a language model. Most of these tools rely on two statistical measures. Perplexity measures how predictable the word choices are. Burstiness measures variation in sentence structure and length.
Language models often produce text with low perplexity and low burstiness. But this is also true for experienced human writers who develop a disciplined style, over years, and sometimes decades of professional writing. Disciplined writing is also common to professionals who draft contracts, pleadings, compliance documents, and academic papers. Clarity, precision and sharp language are characteristics of good legal writing. And these are the identifiers that many AI detection tools interpret as evidence of machine authorship.
The effect is sharper still for writers whose first language is not English. The 2023 research paper published in Patterns found false positive rates of 61% for non-native English speakers performing the same writing tasks compared to 5% for native English speakers. This finding deserves far more attention in India than it has received.
Indian academic writing follows conventions that differ from those commonly found in American or British universities. Indian legal writing similarly relies on a more formal style and structure. Many professionals consciously favour predictable and common phrasing because precision matters more than stylistic variation. The characteristics that institutions encourage, or reward, become the characteristics that commercial AI detectors interpret as suspicious.
This is no longer merely a technical limitation. It becomes a question of procedural fairness once institutions begin attaching consequences to those findings. An AI detection score is not evidence of authorship. It is evidence that a commercial algorithm considers a piece of writing statistically similar to text produced by a language model. Treating that probability as a finding of fact shifts the burden of proof onto the wrong person — Shashank Sharma
AI detection has no legal status
When a student is accused of AI-assisted misconduct, the process that follows is disciplinary, not evidentiary. The institution decides whether misconduct has occurred and the student responds. The AI detection finding remains the foundation of the allegation without any defined procedure allowing its reliability to be questioned.
There is no requirement that a second detection system reach the same conclusion. There is no established right, or even means, to inspect the basis on which the software reached its finding. Most importantly, no Indian court has as yet considered what evidentiary status such a finding should carry.
The international position is starting to shift, but the direction in which it’s moving is far narrower than it first appears. In January 2026, the Supreme Court of New York, Nassau County, ruled in Matter of Newby v. Adelphi University6, annulling a university academic integrity finding against Orion Newby. Newby was a freshman with Level 2 Autism Spectrum Disorder who faced discipline after a Turnitin AI detector flagged his essay. Though Newby maintained he wrote the essay with standard tutor assistance and provided two conflicting human written detection scores, Adelphi upheld the finding. The university failed to produce the Turnitin documentation, ignored his disability accommodations, and let the same administrator act as both accuser and appellate judge. Justice Randy Sue Marber ultimately ruled the university process arbitrary and capricious due to these systemic breakdowns.
Crucially, this distinction is not a technicality because it is the entire point. The court did not rule on the validity of AI detection scores or lay down how much weight they should carry. Instead, it intervened because Adelphi’s internal due process failed so badly that it was indefensible on its own terms. Remove the missing documentation, the conflicted appeal, and the ignored contradictory results, and nothing in Newby guides a university on how to safely rely on an AI detection score. The ruling merely exposes a vacuum regarding AI evidence, it does not fill it.
A separate suit, filed in the U.S. District Court for the District of Connecticut by a French-born Yale Executive MBA student, argues that his formal writing style and status as a non-native English speaker produced a false positive finding against him7. This case has not been decided, but it’s built on the same premise as the research this piece has already cited, that the traits AI detectors treat as suspicious are often just the marks of careful, deliberate writing.
Both cases matter because they show courts starting to scrutinise the detection tools themselves instead of accepting their conclusions at face value. India has not reached that stage. No reported decision here has asked whether a detection score satisfies any recognised evidentiary threshold before it is allowed to shape a disciplinary or professional outcome.
The same absence of standards affects professionals outside academia. An advocate whose article is rejected after an AI detection finding, a researcher whose manuscript is questioned, or a professional whose work product attracts suspicion faces the same practical difficulty. There is no recognised mechanism to challenge the technical basis of the allegation. The dispute concerns software, but the consequences belong entirely to the person it accuses.
India's AI Governance Guidelines released ahead of the AI Impact Summit in Feb’2026 acknowledge that individuals often lack accessible grievance mechanisms for AI-related harms and recommend that organisations establish them. Those recommendations are advisory rather than enforceable. They do not specifically address AI detection systems or the procedural safeguards that should accompany them.
The answer is not to abandon AI detection. It is to recognise its proper legal role. A detection score should begin an inquiry rather than conclude one. — Shashank Sharma
The contrast with regulatory scrutiny elsewhere is hard to ignore. The US Federal Trade Commission (FTC) acted against an AI detection company in 2025 after challenging unsupported claims about the accuracy of its software8. OpenAI withdrew its own AI Classifier after acknowledging its low accuracy rate9. Indian regulators have not yet required vendors marketing AI detection systems to educational institutions or organisations to substantiate the accuracy claims which drive the purchasing decisions.
The answer is not to abandon AI detection. It is to recognise its proper legal role. A detection score should begin an inquiry rather than conclude one. A document's version history, draft timestamps, collaborative editing records, and other contemporaneous material establish authorship far more reliably than a statistical prediction ever can.
Indian courts have long required evidence to satisfy standards of reliability before it influences legal rights. AI detection should not become the exception simply because the underlying mathematics appears sophisticated. Until India decides what evidentiary weight these systems deserve, institutions will continue converting probabilities into findings of fact without a legal framework that justifies doing so.