India Supreme Court voids NCLT ruling over fake AI citations

The integration of Large Language Models (LLMs) into legal research has created a recurring pattern of 'hallucination' incidents where generative AI fabricates plausible but non-existent case law. Prior to this ruling, courts in the United States, United Kingdom, and Canada had issued admonishments and sanctions against lawyers citing fake AI cases, typically treating them as isolated instances of professional negligence or lack of candor. These earlier responses generally focused on punishing individual attorneys rather than establishing systemic evidentiary standards or regulatory frameworks for AI verification. The prevailing judicial approach treated AI hallucinations similarly to traditional citation errors, albeit with heightened scrutiny regarding intent. This incident represents a departure from that pattern by elevating AI fabrication from a disciplinary matter to a jurisdictional defect that voids judicial outcomes. The comparison to industrial toxic releases marks a rhetorical shift, framing AI hallucinations not as mere inaccuracies but as environmental contaminants to the legal ecosystem that require regulatory containment rather than case-by-case correction.

On July 2, 2026, the Supreme Court of India issued a landmark ruling in Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. & Anr. (2026 INSC 668), setting aside judgments from both the National Company Law Tribunal (NCLT) and the National Company Law Appellate Tribunal (NCLAT). According to reports from Bar and Bench and LiveLaw, the apex court vacated these lower tribunal orders after determining they relied upon non-existent legal precedents and fabricated citations generated by artificial intelligence tools. The Supreme Court characterized the reliance on such hallucinated content as "invisible, insidious and catastrophic," comparing the introduction of fake AI-generated judgments into legal proceedings to "the release of methyl isocyanide in the province of law and justice," according to LiveLaw.

The controversy originated when the NCLT issued a ruling that incorporated AI-generated paragraphs falsely attributed to genuine precedents, along with incorrect citations to cases that did not exist. According to a detailed summary by Advocate Mamta Sharma, these errors were not corrected during the appellate review at the NCLAT, allowing the fabricated authorities to persist through two levels of adjudication before reaching the Supreme Court. Upon discovering the reliance on non-existent case law, the Supreme Court held that any judicial decision tainted by even an iota of fake or hallucinated precedent is considered "no decision in the eyes of law" and must be set aside to preserve judicial integrity.

In response to this systemic failure, the Supreme Court established a strict regulatory framework governing AI use in Indian courts. According to Advocate Sharma’s account of the judgment, the Court declared a "zero-tolerance policy" toward citing AI-generated precedents without independent verification. The ruling explicitly defined it as professional misconduct for advocates to cite unverified AI outputs and categorized it as a serious lapse for judges to rely upon them. Furthermore, the Court directed the Bar Council of India (BCI) to examine the issue and frame specific norms to prevent future occurrences, signaling a shift from ad hoc warnings to institutionalized regulation.

Legal news outlets confirmed the immediate regulatory fallout. LiveLaw reported that the Supreme Court specifically tasked the BCI with examining the misuse of AI and establishing examination directives or guidelines for legal practitioners. Bar and Bench corroborated this, noting the Court's mandate for the BCI to frame norms amidst its declaration of zero tolerance. This directive places the responsibility for policing AI verification standards directly on the statutory regulatory body for the legal profession, rather than leaving it to individual judicial discretion.

The ruling has been widely interpreted as establishing that AI tools cannot supplant the fundamental duties of legal professionals. According to Advocate Sharma, the Supreme Court made it "abundantly clear" that while AI may assist the legal profession, it can never replace an advocate's duty of verification or a judge's duty of independent adjudication. This distinction reinforces the principle that technological assistance does not absolve human officers of the court from their ethical and procedural obligations to ensure the authenticity of legal authority.

Multiple sources identified the underlying dispute as related to Essel Infra insolvency proceedings, though the primary cited case name remains Pooja Ramesh Singh. NDTV Profit reported that the Supreme Court quashed an Essel Infra insolvency order specifically due to the tribunals' use of fake AI-generated precedents. This suggests the Pooja Ramesh Singh matter may be the specific vehicle through which the broader insolvency litigation was addressed, or that multiple related matters were consolidated under this scrutiny. Regardless of the specific commercial entity involved, the Supreme Court’s intervention establishes a binding national precedent regarding the admissibility and verification of AI-generated content in Indian jurisprudence.

The sequence of events highlights a critical gap in current appellate oversight. The fact that AI hallucinations survived scrutiny at both the NCLT and NCLAT levels before being caught by the Supreme Court indicates that existing verification protocols were insufficient to detect sophisticated fabrications. By voiding the orders entirely rather than merely correcting the citations, the Supreme Court signaled that the contamination of the judicial record by AI hallucinations is a structural defect that invalidates the entire adjudicative process, not merely a reversible error of fact or law.

The ruling directly affects all advocates and judges operating within Indian tribunals and courts. Advocates now face professional misconduct charges for citing unverified AI precedents, while judges risk having their decisions voided entirely if AI hallucinations contaminate the record. The Bar Council of India must now allocate resources to develop and enforce AI verification norms. For litigants, this creates potential grounds to challenge adverse judgments where AI reliance is suspected, increasing appellate volume. The magnitude extends beyond the specific Essel Infra/Pooja Singh matter to establish nationwide compliance requirements for AI-assisted legal research, effectively making human verification of AI outputs a mandatory procedural step rather than a best practice.