Artificial intelligence can search, summarise, compare and draft, and courts have legitimate uses for those capabilities. As the technology takes on more demanding intellectual work, however, the responsibility for deciding a case must remain with the adjudicator.
The Supreme Court of India addressed that boundary in Pooja Ramesh Singh v Jammu and Kashmir Bank Ltd. & Anr., decided on 2 July 2026. The case concerned an insolvency dispute in which the National Company Law Tribunal had relied on fictitious authorities and fabricated passages attributed to genuine judgments.
The National Company Law Appellate Tribunal referred to the same authorities without detecting the problem. The Supreme Court set both decisions aside and returned the matter for fresh adjudication, without expressing a view on its merits.
One detail makes the case particularly troubling. According to an affidavit filed by the bank, its counsel had not cited the offending authorities; the tribunal had obtained them through its own research. The failure therefore reached into the tribunal’s preparation of its decision and survived an appeal.
The Supreme Court’s response addressed both the reliability of legal sources and the need to retain human control over adjudication. (Judgment, paragraphs 1 and 14–19)
Why fabricated authorities undermine a judgment
A fabricated citation becomes more serious when it enters a judgment. Once a court treats it as established law, an unverified machine-generated proposition has become part of the reasoning by which someone’s rights are determined. Its apparent authority comes from the court’s acceptance of it, making the failure consequential for everyone who relies on that decision.
The Supreme Court treated this as a threat to the integrity of adjudication. It held that a decision founded on fake or hallucinated material must be set aside, irrespective of whether that material had a direct or indirect bearing on the decision-making. The Court expressly preserved the rightful use of AI while rejecting reliance on fabricated material presented as judicial precedent. (Judgment, paragraphs 7 and 17)
The significance is that the integrity of the process matters independently of the result. A judgment exercises public power over the parties before the court, and its reasons must withstand scrutiny. Even an outcome that could be justified on other grounds cannot make a fictional authority a legitimate part of that reasoning.
The problem also affects the trust on which legal proceedings depend. Courts ordinarily rely on advocates to cite genuine authorities accurately. If every citation becomes suspect, the work of establishing what the law says becomes slower and more burdensome.
In this case, the failure of the appellate process to detect the fabricated material showed how an error could pass through successive levels of scrutiny.
The risk of habitual dependence
The judgment’s concern extended beyond inaccurate output to the habits that AI may encourage. A system that performs difficult work quickly makes it tempting to delegate more of that work, especially under pressure. The danger grows when the professional begins accepting a plausible answer without doing the intellectual work needed to assess it.
That dependence can be difficult to recognise because the output may look like competent legal reasoning. It can identify an issue, state a rule and offer a conclusion in familiar language. Reading such an answer and finding it persuasive can feel like having reasoned through the problem, even when its assumptions, omissions and sources remain unexamined.
Judicial decision-making requires the adjudicator to assess evidence, determine which facts have been established, interpret the applicable law and reconcile competing authorities.
Difficult cases may also require choices between competing legal and social values. AI’s ability to contribute to these tasks does not relieve the judge of the obligation to understand and justify those choices.
This is why the Court’s insistence on “a human in the loop at every stage” matters. In practice, meaningful oversight requires more than a judge reading and signing a completed draft. The adjudicator must remain capable of questioning the analysis, rejecting its premises and explaining the conclusion in light of the evidence and law. (Judgment, paragraphs 1–5)
Verification is part of the legal work
For advocates, the immediate obligation is familiar: an authority must be checked before it is cited. The Supreme Court characterised citing AI-generated judgments without verification as misconduct and directed the Bar Council of India to constitute a committee to develop guidance and disciplinary consequences. (Judgment, paragraphs 7–9)
Verification involves more than confirming that a case name appears in a database. The lawyer must read the judgment, check the quoted passage, establish whether the decision remains good law and determine whether it supports the proposition being advanced. A real case can still be misrepresented by an invented quotation or a summary that omits the qualification on which the reasoning depends.
The same discipline applies when AI is used to summarise a record or test an argument. A useful summary may save time, but the lawyer still needs enough knowledge of the underlying material to recognise what has been lost or distorted. A suggested weakness in an argument deserves examination before it shapes advice to a client or submissions to a court.
AI can increase the amount of material a lawyer examines and reduce the time spent on preliminary work. Those gains have professional value when they leave the lawyer better informed and able to defend the resulting advice. They become a liability when the lawyer can no longer explain the work without referring back to the system that produced it.
Keeping courts accountable
For judicial institutions, the practical task is to decide where AI assistance is appropriate and how its contribution will be checked. Human verification belongs within the research and reasoning process, where errors can still be challenged before they shape the decision. A final review is of limited value if the reviewer has already accepted the system’s account of the facts and law.
For Tanzanian lawyers and courts considering AI, this Indian judgment offers a useful starting point for that discussion. The questions concern who checks the sources, who evaluates the reasoning and who takes responsibility when the work is adopted. Those responsibilities need to remain clear as the technology becomes more capable.
Search tools and legal databases have long helped practitioners locate information. Generative AI can go further by producing the apparent substance of an argument, which makes the temptation to delegate greater. A court using that assistance must still be able to explain why the evidence and the law justify its conclusion.
The people whose rights are being determined deserve reasons the adjudicator has examined and can defend.
_____________
LawCraft Attorneys advises businesses on regulatory compliance, commercial litigation and dispute resolution across Tanzania. For guidance on a specific matter, contact info@lawcraft.co.tz or call +255 744 48 63 64.
