Phantom Precedents: How AI Hallucinations Are Getting Indian Lawyers Sanctioned
L
LawWorld
23 Jul 2026
A citation looks perfect. It carries a proper SCC or AIR reference, a plausible bench, a confident holding that supports your argument exactly. There is only one problem: the case does not exist. Welcome to the era of the phantom precedent — and it is quietly becoming one of the most dangerous professional risks facing Indian advocates.
This is not scaremongering. It is a documented, accelerating pattern in Indian courtrooms. Understanding how it happens — and how to make it impossible in your own practice — is now part of basic professional competence.
What an AI hallucination actually is
When a general-purpose AI model generates text, it is predicting the most statistically likely words to follow your prompt. It is not consulting a database of real judgments. So when you ask for "a Supreme Court case supporting X," a model will happily produce a case name, a citation, and a summary that sound right — because sounding right is exactly what it was trained to do. The Supreme Court's own Draft AI Regulations of 2026 gave this a name: AI hallucination, defined as the generation of fictitious or inaccurate legal content, including fabricated precedents, statutory provisions, and quotations.
The output is dangerous precisely because it is fluent. There is no spelling mistake, no obvious tell. A busy advocate under deadline pressure sees a well-formatted citation and assumes it is real.
The cases Indian courts are already catching
The pattern is no longer theoretical:
Bombay High Court (January 2026): A party was slapped with costs of ₹50,000 for dumping fabricated case laws into written submissions. The judge noted tell-tale signs of raw AI output, including formatting artefacts that no genuine research would produce.
Delhi High Court (September 2025): In Greenopolis Welfare Association v. Narender Singh, a petition was withdrawn in embarrassment after opposing counsel exposed the citations as fabricated — including invented paragraphs 73 and 74 of a landmark judgment that contains only 27 paragraphs.
Karnataka High Court (March 2025): The Court directed an inquiry against a trial judge who had relied on two non-existent Supreme Court decisions to reject an application — a rare instance of an appellate court moving against a judge, not just an advocate.
Bombay High Court (October 2025): A faceless tax assessment order of roughly ₹27.91 crore was quashed after it was found to rest on three non-existent precedents.
Globally, researchers tracking this phenomenon have logged well over a thousand instances and counting, with the rate rising from a couple per week to several per day through 2025. India is firmly on that curve.
Why smart, experienced lawyers still fall for it
Three forces combine. Time pressure: the temptation to trust a fast answer is strongest exactly when you can least afford to be wrong. Fluency bias: we are wired to trust confident, well-structured prose. Tool mismatch: using a consumer chatbot for legal authority is like using a weather app for surgery — impressive technology, wrong instrument.
Learning to spot the tell-tale signs
Even before verification, some fabrications carry warning signs. Citations that are suspiciously perfect for your argument, holdings phrased in unusually neat language, paragraph numbers that seem too high for a short judgment, and formatting quirks like repetitive structure or stray markers can all indicate raw AI output. In the Bombay High Court matter, the judge specifically pointed to formatting artefacts as evidence the submission came straight from a model. Train your eye to treat "too good to be true" citations with extra suspicion — and then verify regardless, because the most dangerous hallucinations are the ones that look entirely ordinary.
The workflow that makes hallucinations impossible
You do not have to choose between speed and safety. You have to choose the right tool and the right habit.
Use retrieval-grounded legal AI, not open chatbots. The core fix is architectural. A platform that retrieves from an actual corpus of Indian judgments — and shows you the source document behind every citation — cannot invent a case, because it can only surface cases that exist. When the citation links back to a real judgment you can open, the phantom problem disappears.
Adopt a two-minute verification ritual. For every authority you intend to cite: open the judgment, confirm the citation, locate the exact paragraph, and read enough to confirm it says what you think it says. If you cannot find it, do not cite it. This single habit would have prevented every case listed above.
Never let AI be the last step. AI compresses research; your professional judgment closes it. The advocate signs the filing — and the advocate, not the algorithm, answers for it.
The takeaway
AI is one of the most powerful productivity tools ever handed to the legal profession. It can turn hours of research into minutes and free you to do the work only a lawyer can do — think, argue, advise. But that power comes with a non-negotiable condition: verification. The advocates who thrive in this era will be the ones who use AI aggressively for speed and rigorously for accuracy.
Phantom precedents only haunt those who cite what they have not read. Ground your research in real, linkable Indian law — the design principle behind legal-first platforms like LawWorld — and the ghosts have nowhere to hide.
This article is for general information and does not constitute legal advice.
12% read
Keep reading — it’s free.
Create a free LawWorld account to finish this piece and unlock the full library of practice guides written for advocates in India.
Finish every guide, free
Save articles to read later
New practice insights weekly
Free forever. No card required.
Comments
Sign in to read the discussion and share your thoughts.
