Ask a chatbot a question today and it answers at once, fluent and sure of itself. That confidence is the problem. Sometimes the answer is simply made up, a fact, a quote, a court case, delivered in the same calm voice as the truth. Most of the time this costs us nothing. In a courtroom it can cost someone their case, because a judge may rule on something that was never real.
Lawyers use these tools now, the same as everyone else. They draft with AI, summarise with AI, and look up old judgments with AI. Supreme Court of India has gone further than most. Where courts elsewhere, in England, in Singapore, have offered judges advisory guidance on AI, the Supreme Court of India has proposed something binding and comprehensive, covering every court and tribunal in the country. That makes it one of the first courts anywhere to move from suggestion to rule.
Much of the draft is good. The human judge stays in charge. Nobody gets to shrug and blame the machine. It deserves the praise it has drawn.
One rule, though, does not sit right. The rule on disclosure. It says this: if a lawyer uses an AI tool to prepare a document, the lawyer must tell the court. Fair enough, on the face of it.
But picture a student with a calculator. She can say, quite truthfully, that she used one. That tells you nothing about whether her answer is correct. Press the wrong key and the answer is still wrong, calculator declared or not. Owning up to Representational Image. the tool and getting the sum right are two separate things.
Disclosure is the same. Telling the court you used AI says nothing about whether what it gave you is true. And truth is the whole point. The danger was never that lawyers use AI. The danger is that AI invents things and says them without a flicker of doubt.
We do not have to imagine it. It has already happened, here. In Gummadi Usha Rani v Sure Mallikarjuna Rao, a trial court in Andhra Pradesh decided a case on the strength of four earlier judgments. The four did not exist. They had names, numbers, years, everything that makes a citation look real, and not one of them had ever been decided. The AI had dreamt them up, and they walked straight into a court order.
The Supreme Court, when it saw this, did not call it a slip. It said relying on fake, AI-made judgments can amount to misconduct.
Is a disclosure rule enough to prevent all of this? No. Even if the judge had honestly ticked the ‘yes’ box for using AI, the four fake cases would have been added to the order. The reason is that honesty about using AI, and honesty about citing correct law, are two different things.
This rule simply asks: Were you truthful when you said you used AI? It does not ask: Are you certain the answers you got by using AI are correct?
There is another example. A lawyer in New York submitted a brief based upon six fabricated case citations provided by ChatGPT. Prior to submitting his brief, he performed a task that most of us perform prior to filing anything. He asked ChatGPT if those cases existed. It answered that they did exist. They did not. The Judge imposed penalties on the attorneys and referred to their arguments as nonsense.
His mistake was not using AI, his mistake was relying on its accuracy without checking it first.
Thus, there is a void in the current draft. Disclosures identify which tools were used. However, disclosures do not determine if the results produced by the tools are accurate. Instead of addressing the issue disclosed through disclosure, the problem remains unsolved.
Thus, if a pleading contains references to specific judicial decisions or factual evidence, attorneys should confirm that each citation contained in the pleading corresponds with actual databases available online (i.e. SCC Online, Manupatra or Westlaw).
It is not necessary to eliminate disclosures entirely. What is needed is to include the missing portion of the disclosure requirements.
A good teacher does not simply ask if a student used a calculator. She asks students to provide proof of how they arrived at their final solution. The rules governing legal representation should do the same.
Not only should advocates disclose whether AI was used to prepare pleadings, etc., but also demonstrate what steps they took to verify the information generated by AI. When the consequences of false or inaccurate data are serious (such as during criminal proceedings), attorneys should be required to demonstrate more than just a cursory review.
Similarly, if an advocate claims that a decision cited in a pleading states exactly what she claims it states, then she should confirm her assertion through reference to a reliable database.
It is not burdensome for advocates to follow these procedures. As careful professionals, advocates already check references and cite authority before submitting pleadings. Therefore, the only additional requirement under this procedure would be to require attorneys to document that factually verifiable steps have been taken to confirm the accuracy of their submissions.
Therefore, it is not excessive for courts to continue building on their efforts by implementing verification procedures along with disclosure mechanisms.
Nevertheless, if courts merely adopt disclosure requirements, while ignoring the need for verification, advocates will be able to submit pleadings containing false and misleading statements regarding the sources of their contentions and judges will continue to allow them into court proceedings.
The Court has built something worth having, and it is right to move before others do. But a rule that asks only whether AI was used tells us the tool was in the room. It does not tell us the work is sound.
If disclosure is meant to guard the honesty of the record, and not merely to keep a register of who used a machine, the draft must take the last step and tie disclosure to verification. Leave that step out, and we are left with a form everyone signs honestly, and fake cases still finding their way into our judgments.
Professor (Dr.) Aneesha Shams is Dean, School of Law, IILM University, Gurugram Dr. Tushar V Sharma is Director, AI Policy Lab & Assistant Professor, School of Law, GD Goenka University. The views are personal.
