The Supreme Court's recent stance on AI in legal practice has stirred up plenty of conversation in the profession, and rightly so. The concerns are real: fabricated citations, algorithmic bias, privacy risks, and the fear of machines creeping into adjudication. The draft regulations make one thing clear. AI must remain strictly assistive, and the burden of accuracy sits squarely on the lawyer or judge using it.
I have no quarrel with that principle. In fact, I would argue it is exactly the right one. But having spent years building and working with these tools, I want to offer a practitioner's perspective on each of these concerns, because I believe most of them say more about how we use AI than about AI itself.
Start with what AI actually is
Strip away the hype, and AI is a tool, a medium entirely managed by its user. It does not create anything on its own. A large language model is trained on vast amounts of information, and when you prompt it, it draws on that training to produce a response. That is it. There is no independent will, no agenda, no mind of its own.
Which means the shocks, surprises, and inaccuracies we keep hearing about do not appear out of thin air. They creep in when the tool is managed badly: when the underlying data is limited, or when the prompting is careless. A poorly framed question gets a poorly grounded answer. That is not a revelation about artificial intelligence; it is true of every instrument a professional has ever used.
A symptom, not a disease
The problem of fake citations is the judiciary's loudest concern, and the recent NCLT episode, where fabricated cases found their way into a filing, shows why. The Court has warned, in no uncertain terms, that citing invented AI material pollutes the legal system and will be penalized.
Fair enough. But let us be clear about what a hallucination actually is. It happens when the model tries to deliver a response without having the right information, often because the user gave it no context, no grounding, no source material to work from. Ask a general purpose chatbot for a precedent it has never seen, and it will do what it was built to do: generate text that sounds plausible. The failure is not malice; it is a tool being pushed beyond what it was given.
An advocate who cites a case without reading it has always been negligent, long before AI existed.The duty never changed
The fix is equally clear. Ground your research in verified sources, provide proper context, and verify every citation before it goes anywhere near a filing. The technology has not changed that duty; it has only made shortcuts more tempting.
On adjudication, the Court has it exactly right
Here I will simply agree with the judiciary. Judges should not use AI for decisions, sentencing, or predicting outcomes. Justice demands empathy, moral reasoning, and an understanding of human context that no model possesses.
But this was never a real tension, because AI's genuine value in a courtroom lies elsewhere. It is a way to access, organize, format, and deliver information: an analytical assistant, not an oracle. Let it summarize voluminous records, translate proceedings, surface relevant precedents, and prepare case briefs. The judge decides. There is no need to make the tool the decision maker, and frankly, nobody serious is asking for that.
The one concern that demands immediate discipline
If there is a concern I would underline twice, it is this one. Court records and case files carry deeply sensitive personal information, and pasting details of an active matter into an open, public AI chat window is genuinely reckless. Once that data leaves your hands, you have no control over where it goes.
The answer is not to abandon AI. It is to choose your tools like a professional. Use secure platforms built for legal work, with proper data protections, and there are good ones available today. Keep details of active matters out of consumer chatbots entirely. This is basic professional hygiene, no different from not discussing a client's case loudly in a cafe.
The real answer: legal temper, not legal panic
Every advocate develops what I would call a legal temper, a professional maturity built during training on campus and sharpened on the job. It is the instinct that tells you which source to trust, which argument to press, which shortcut to refuse.
That same temper is what is needed with AI. These tools reward judgment and punish laziness. A mature practitioner who verifies, checks thoroughly, and applies independent reasoning will find AI an extraordinary force multiplier. A careless one will find trouble, just as they always have, with or without technology.
Mandatory disclosure: a phase, not a permanent fixture
The draft regulations require lawyers to disclose when AI was used to prepare documents or arguments, with full accountability resting on the advocate if anything fabricated slips through.
Accountability, absolutely. That was never in question. The advocate signs the document; the advocate owns it. But the disclosure requirement itself strikes me as a temporary measure, born of this early, anxious moment. AI is a tool, like a PC. It is not responsible for errors or omissions; the person using it is. Requiring a certificate for drafting assisted by AI is a bit like demanding that every document typed on a computer be certified as such. As the profession matures in its use of these tools, I suspect this requirement will quietly fade.
The opportunity we should not lose sight of
Amid all this caution, let us not lose the bigger picture. Modern AI systems, and Claude for instance is remarkably well versed in Indian jurisprudence, can make a diligent advocate dramatically more efficient. Research that took days now takes hours. Drafting that consumed an evening takes an afternoon. The time saved simply cannot be underplayed.
And that matters enormously in India, where crores of cases sit pending across our courts. Used judiciously, with grounding, verification, and secure tools, AI may well prove to be the answer to that seemingly insurmountable pendency. Not by replacing judges or lawyers, but by clearing the mechanical burden that buries them, so that human judgment can be spent where it actually belongs.