Why Indian Advocates Cite Cases They Have Not Read

A tax tribunal in Bengaluru recalled an order worth roughly ₹669 crore after four of the authorities it rested on turned out not to exist. The Bombay High Court quashed an assessment of nearly ₹28 crore built on three invented precedents.¹ In July the Supreme Court held that citing a hallucinated judgment without verifying it is misconduct on the part of an advocate, and that an order resting on such material is a nullity.²

Every account of this has treated it as a story about artificial intelligence. It is not. The machine supplied a new way of doing something the profession has been doing, and being told off for doing, since well before any of the advocates involved were born.

The complaint is fifty-eight years old

In November 1967, in State of Orissa v. Sudhansu Sekhar Misra, 1967 INSC 252, the Supreme Court said this:

“A decision is only an authority for what it actually decides. What is of the essence in a decision is its ratio and not every observation found therein nor what logically follows from the various observations made in it.”³

That is not a remark about sloppy drafting. It is a description of a specific professional habit: counsel finding a sentence that suits them, and presenting it as though the Court had decided the point.

The Court said it again in 2002, in Padma Sundara Rao, warning that courts should not rely on a decision without examining how its facts fit the case at hand, because “one additional or different fact may make a world of difference between conclusions in two cases.”⁴

And again in 2004, in Bharat Petroleum Corporation Ltd. v. N.R. Vairamani, 2004 INSC 569, in the passage every law student is made to copy out and almost nobody applies:

“Courts should not place reliance on decisions without discussing as to how the factual situation fits in with the fact situation of the decision on which reliance is placed. Observations of Courts are neither to be read as Euclid’s theorems nor as provisions of the statute and that too taken out of their context… Judgments of Courts are not to be construed as statutes… Judges interpret statutes, they do not interpret judgments.”⁵

1967. 2002. 2004. 2026. The same complaint, in almost the same words, across fifty-eight years and four generations of counsel. A problem that recurs that consistently is not a problem about technology. It is a problem about what the profession considers adequate.

What citing actually involves, in practice

Consider what a junior is usually asked to do. Find me something on this point. The realistic sequence that follows is: search a database for a phrase, open the results, read the headnote, find one that says what is needed, copy the citation into the written submissions, move on.

Notice what has not happened. Nobody has read the judgment. Nobody has checked the facts it was decided on. Nobody has established that the sentence relied on was necessary to the decision rather than an observation made in passing. Nobody has checked whether it has since been overruled, distinguished or referred to a larger Bench.

The headnote deserves particular blame here. It is not written by the Court. It is written by the publisher’s editors, it carries no authority whatsoever, and it is a finding aid that has been quietly promoted into a substitute for reading. A great deal of Indian legal argument is, in practice, argument about headnotes conducted by people who believe they are arguing about judgments.

Why nothing ever forced a change

Here is the part that explains the fifty-eight years.

If you cite a real case for a proposition it does not support, almost nothing happens. The case exists. The citation is correct. The sentence you quoted is genuinely in the judgment. To expose you, the other side must read the whole judgment, work out what it actually decided, and persuade the court that your reading is wrong. And they are usually working under the same time pressure and the same habits that produced your error. Most of the time nobody checks. When somebody does, the worst outcome is that your point is not accepted. It looks like losing an argument, not like a failure of professional duty.

The failure, in other words, has always been invisible. It leaves no trace that anyone can point to afterwards.

A hallucinated citation is different in exactly one respect. It is falsifiable in seconds. Anyone can type the case name into a database and watch nothing come back. For the first time, the gap between what an advocate asserted and what an advocate verified is visible on the face of the record, to a judge, without anyone having to read anything.

That is why this is being treated as a scandal and the older version never was. Not because it is worse. Because it can be seen.

The equivalence nobody wants to state

An advocate who cites a case that does not exist and an advocate who cites a real case for something it did not decide are doing the same thing. Both are asserting to a court that an authority supports a proposition, without having verified that it does. Both are asking a judge to rely on a representation they have not checked.

The only difference is that one of them can be caught by a database search.

I am not suggesting the Supreme Court was wrong to act, or that the two are identical in culpability: a fabricated authority carries an element of pure invention that a misread one does not. But if the wrong being punished is citing without verifying, then the profession should be honest that it has been committing that wrong at scale, in every court in the country, for as long as anyone can remember, and that the machine merely produced a version of it that could be detected.

What the new rules will and will not fix

The consequence of the AI rulings will be that advocates start checking that the cases they cite exist. That is worth having. It is also the smallest possible part of the problem.

Nothing in the current response requires anyone to check what a case held. A junior who confirms that every citation in a set of submissions is real, and confirms nothing else, will have complied fully with the standard now being enforced, and will still be capable of misleading a court on every point in the list.

If the profession wanted to fix the actual defect, the standard would not be “the case exists.” It would be closer to this: for every authority relied on, you can state in one sentence what it decided, identify the paragraph you are relying on, say whether that passage was necessary to the decision, and confirm it has not since been overruled. That is not an onerous test. It is what Sudhansu Sekhar Misra assumed counsel were doing in 1967.

Law schools could require it in every moot memorial. Chambers could require it of every junior. Neither needs a rule from anybody to start.

What follows

The uncomfortable thing about the hallucination cases is not what they reveal about the software. It is what they reveal about the checking. A tool that invents citations only causes damage in a system where nobody was verifying citations in the first place. In a profession that read what it cited, a fabricated case would have died at the first reading, in chambers, and never reached a tribunal at all.

It reached a tribunal, and cost ₹669 crore, because it was travelling through a process that had no step in it for reading. That step was missing long before the machine arrived, and removing the machine will not put it back.

Filed under The Legal Profession