For the last two years, stories about AI hallucinations in court have usually followed the same pattern. A lawyer files a brief containing a fabricated case, opposing counsel or the judge catches it, sanctions follow, and the legal press gets another cautionary tale.
In that story, the role of judge is to catch what the lawyer missed.
A judgment handed down by the High Court of Jammu & Kashmir and Ladakh on 6 June 2026 quietly flips that narrative. In Woodland House School v. Malik (CM(M) 191/2026), the questionable citations were not in a pleading. They were in a judicial order.
What Happened
The underlying dispute is fairly boring: a long-running employment case between a Srinagar school and a former employee, an interim order directing partial salary payments, and years of what the High Court described as deliberate obstruction by the school. The petition was dismissed with costs, and the execution order under challenge was upheld.
The interesting part comes in a postscript.
Counsel for the school argued that the trial court's order relied heavily on precedents that were incompletely or inaccurately cited, without clearly identifying their ratio or explaining their relevance to the facts.
The High Court independently verified the authorities cited in the trial court's order. The results will sound familiar to anyone who has reviewed AI-assisted legal drafting.
Of the two principal authorities relied upon, the first carried both an incorrect citation and a case title that could not be traced. In the court's words, the judgment "does not appear to exist" in the form cited. The second referred to a genuine case, but the citation pointed somewhere else entirely.
These are two of the most common citation failures in AI-generated legal text and exactly the kinds of issues verification systems are designed to catch.
What the Court Didn't Say
It's important to be precise here. The High Court never found that the trial judge used an AI tool. It could not have. There is no disclosure requirement, no audit trail, and nothing in a finished judgment that reveals how it was drafted.
Once AI-assisted drafting becomes commonplace, every unverifiable citation invites the same question.
What is notable is what came next. After describing the untraceable and mismatched citations, the court turned to the growing use of AI in judicial work and issued guidance for judicial officers across the Union Territory. It set out a verification protocol:
Any citation obtained through AI tools must be verified against authentic sources before appearing in an order. Every precedent must carry a complete and accurate citation. Where a precedent forms the basis of a finding, the relevant passage should be reproduced verbatim. Citations from unofficial databases must be cross-verified before use.
The court was equally clear about responsibility: the authoring judge remains accountable for every authority appearing in an order. Tools may assist; they do not absolve.
Why This Matters More Than the Lawyer Cases
When a lawyer files a hallucinated brief, the common law legal system has a built-in corrective mechanism. There is an opposing party motivated to find the error, and a judge obliged to rule on it.
When the hallucination appears in a judicial order, those safeguards largely disappear. In Woodland House School, the issue surfaced only because a litigant pressed the point on appeal and a High Court judge personally re-ran the research. Most trial court orders will never receive that level of scrutiny.
A hallucinated citation can sit inside an otherwise correct decision, unnoticed for years, causing little practical harm to the result but eroding trust in the process.
What is equally striking is what the High Court did with the discovery. It declined to disturb the order. The underlying conclusion was sound, and the petitioner's conduct had been abusive regardless of the citation problems.
The bad citations changed nothing about the outcome. They changed a great deal about confidence in the court's paper trail. That is the uncomfortable lesson.
The ambiguity cuts the other way as well. Because the court could not establish how the citations appeared, a simple human mistake is now difficult to distinguish from machine fabrication. Once AI-assisted drafting becomes commonplace, every unverifiable citation invites the same question.
Is this a one off? Or will this be a feature going forward?
The Srinagar guidance is close to a model policy, and notably more concrete than many judicial AI discussions elsewhere. But read the directives as an operational requirement. They effectively ask every judicial officer to trace every cited authority to an authentic source, confirm the citation matches the case, identify the relevant passage, and verify that the proposition attributed to the case is actually supported by the text.
For every authority. In every order.
That is exactly the work the High Court performed here. It is also a huge burden of work that does not scale well when done manually.
Verification at Scale
So that begs the question, how will our legal systems cope with the AI revolution? Much has been written about the issues and opportunities facing private practice legal professials but less has been said about the impact on the Judiciary. We did not design our products with judges in mind, however it seems that they may also benefit from more reliable AI.
Woodland House School v. Malik will ultimately be remembered less for its holding than for its postscript: the moment a judge made up citations and an Indian High Court treated citation verification as a matter of judicial legitimacy.
AI drafting tools are clearly already in use in judges' chambers as much as in lawyers' offices. The question is no longer whether AI will assist legal drafting. It is whether verification happens by default, or only when someone appeals.
If your chambers, registry, or legal team is working out how to meet that standard at scale, we'd be glad to show you what real-time verification looks like on your own drafts.