As on 19th August 2025, an order was passed by the trial judge in Vijayawada disposing objections to report of a court-appointed commissioner, wherein four supreme court judgements where cited. The issue being the aforementioned judgements never really existed, rather they had been generated via an AI tool, and subsequently pasted into the order without due diligence or verification. On being transferred to the Andhra Pradesh High Court, the bench acknowledged fabrications within the judgment, however the revision petition had been dismissed on merits by relying upon Gummadi Usha Rani v Sure Malikarjuna Rao, as the primary authority for judicial intervention on artificial intelligence in Indian legal practice.
On 27th February 2026, Supreme Court deliberated over the same vide the dual bench of Justices P.S. Narasimha and Alok Aradhe, wherein they declined to treat trial court’s reliance on fake citations as mere lapse in adjudicatory authorities. As per the order, a decision based on non-existent and fake judgements cannot be chalked up to mere decision making error, instead it falls within the ambit of misconduct and subsequent legal consequences must follow. Thus, a notice was issued to Attorney General, the Solicitor General, and the Bar Council of India; Senior Advocate Shyam Divan was appointed amicus curiae. The matter is currently pending for further hearing before the Supreme Court, wherein due instructions are to be passed on 26th May 2026.
The main objective of this paper is to accurately determine allocation of liability with regards to aforementioned hallucinations. Though the court’s framing is sound doctrinally in allocating the liability to the signing advocate, the existing framework remains under protective on two grounds; i. lacking supervisory liability for senior advocates and chambers, and ii. Lacking vendor liability for legal AI products which have been marketed as being ‘hallucination free’, and the paper, thereby attempts to address the aforementioned issues.
- Indian Framework
The Indian response with regards to the aforementioned issue, is somewhat a patchwork. The Andhra Pradesh HC decision did not invalidate an order wherein the reasoning provided by the trial court was doctrinally sound. The cited authorities were treated as mere cosmetics, thereby invalidating their value as binding precedents. The SC’s characterization of the conduct as misconduct rather than mere error, is in part a sound decision.
It would be incorrect to state the pattern is confined to the state of Andhra Pradesh. In the recent past, Bombay High Court has also imposed costs on litigants for filing AI-hallucinated submissions. Similarly, the Delhi High Court had also dismissed petitions wherein the opposing counsel were found to have fabricated citations; The Income Tax Appellate Tribunal at Bengaluru recalled an order which was allegedly traced to unverified use of ChatGPT. Thus, the aforementioned instances vested Supreme Court with the due authority to characterize the conduct as misconduct.
The AI Governance Guidelines released by India in February 2026, was for the sole purpose of reviewing the regulatory gaps comprehensively without imposing sector-specific mandates, and the Supreme Court’s Centre for Research and Planning had published a White Paper in November 2025 alerting to the dangers of hallucination and highlighting instances of AI-generated content in trials that contained non-existent precedents.
They all stem from the same place in the chain and that is the person signing the advocates: the person who decides. This is the beginning you should go with. That’s not the whole story.
- Whether Advocate Liability is the right starting point?
There are three doctrinal aspects which carry the primary responsibility of the individual advocate. The duty of candour and the prohibition against misleading the court are set out in the Standards of Professional Conduct of the Bar Council of India (B.B.I.). Pleadings under Order VI Rule 15 of the C.C.P must be verified and the attorney’s signature is presumed to be a personal guarantee of what is presented to the court.
No obligation to the chamber, the firm or the tool, in any developed sense, but all obligations to the individual advocate. The advocate there had been convicted on the basis of a fake order setting aside his previous conviction, and the Supreme Court there had taken it very seriously and imposed a permanent suspension on him, which was doubled to two years by the Bar Council. What becomes clear is the principle that the remedy for fraud on the court is based on the degree of systemic misrepresentation, not on whether the individual advocate is the perpetrator of the misrepresentation or not. The forgery in Hikmat Ali v Ishwar Prasad Arya was made by another person, and the advocate’s responsibility was to present it before the court to test the means of authenticity. What the advocate did was to file an “AI-generated” citation, and that does not excuse the advocate from the responsibility. The court’s reliance on the filed material, and the systemic harm resulting from that reliance, are the same in both cases. In R. Muthukrishnan v. Registrar General, Madras High Court reiterated that advocate’s duty of candour is towards the court and not the working procedure which led to the misrepresentation in filing.
The Supreme Court’s framing is thus a restoration of doctrine, not an innovation, because it’s not based on error but on misconduct. Any prescription plan should be based on the signature-based approach, and not replace it. However, there are two contexts in which an advocate-driven system has its weaknesses.
- The Two Gaps
- Supervisory Liability
The duties of the senior lawyer to supervise the conduct of junior lawyers or law clerks have no counterpart in Indian professional responsibility law; Section 35 can only reach the senior lawyer in this situation indirectly and poorly through the vicarious liability in tort or through the partnership doctrine, the latter of which is not developed for legal practice.
The obvious objection is that the structure of the chamber in India is different from the structure of the American firms, the junior’s chamber is an informal association of the juniors, who are not in an employment relationship with the senior. A duty commissioned for a hierarchical company might not be easily mapped. There are two possible answers. Firstly, the rule is intended to govern the supervisory role that the senior assumes over a junior, and that is what the rule is about, not formal employment. Second, the policy issue is the same: AI uptake will happen at the junior and clerk level, and a policy that only penalizes the signatory will encourage seniors to deliberately avoid making themselves available to draft, a principle that would drive them to be more and more distant from the actual work. The ABA’s Formal Opinion 512 (2024) adopted Rule 5.3 and acknowledged exactly this situation, but the rule does not have to be taken as a whole. The recognition of the fact that supervisory failure is misconduct and the rule making power of the BCI in Section 49 is the correct instrument is what is needed to be transplanted.
- Vendor Accountability
In India, an apt representation of Legal-AI vendors would be BharatLaw. Products with explicit accuracy representations: AI, Manupatra’s AI products, SCC Online’s AI tools and the Supreme Court’s SUPACE. If an advocate is relying on such a representation and the product goes on hallucinating, there’s a meaningful avenue to the vendor that can be pursued by Indian cause of action. Theoretically at least, there are Acts of the legislature that provide protection for consumers: the Consumer Protection Act, 2019, the Sale of Goods Act and the tort of negligent misrepresentation – but these are not used in practice, and vendor disclaimers are intended to prevent that.
Strict liability would destroy the legal-AI market or push it overseas and ordinary negligence becomes a form of the existing disclaimer regime. Where a vendor specifically states that its product references verified citations, points to a curated database of real citations, otherwise warrants accuracy in information beyond what a typical consumer might receive from a general-purpose generative AI tool, or any other way makes a statement about its product that goes beyond what a typical consumer would receive from a general-purpose generative AI tool, a statement should be actionable by a professional user who relied on it. Disciplinary costs and reputational remediation should be a part of the damages, not just the licence fee. The responsibility for proving reasonable reliance should be on the advocate and the responsibility for proving that the marketing claim is substantiated by independent audit should be on the vendor. A distinction which the law of misrepresentation already recognises and which should not be used to immunise a party against liability for any positive representations they have made expressly in the marketing of goods to a professional user. It is, in substance, Hedley Byrne transposed to a regulated profession, the statutory anchor being the provisions of the Consumer Protection Act on unfair trade practice and deficiency of service.
That’s a real regulatory hole. Both gaps are structural in nature and can be rectified by reform efforts on existing institutional structures, the first being synthetic information labelling on platforms and the second being AI in legal research.
- A Way Forward
Three changes, applied tightly, would seal the holes without tarnishing the signature-based framework. The Bar Council of India must first amend its Standards of Professional Conduct to include an express supervisory duty: Senior advocates and chamber heads must make sure that any document leaving the chamber is accompanied by a check that the author has checked the authority for that document, as it was generated by AI. Second, the AI Governance Guidelines should be translated into industry-specific guidelines for legal-AI providers, which will include disclosure requirements regarding the risk of hallucination, a regime of penalty for inflated marketing claims that are not substantiated, and a calibrated regime of negligent-misrepresentation liability as described above. Third, pleading rules ought to be modified to mandate a brief certificate of use of AI; akin to the Order VI Rules verification, that certifies that any material used has been independently verified.
The natural counter-argument is that such changes will introduce unnecessary impediments in a cumbersome system. The objections are too low a bar for what’s at stake and the design of the proposals. There is no requirement to file anything because of the supervisory duty – it simply spells out which existing actor is responsible for the failures of verification that are happening. The vendor rule is simply a rule that is outside of the docket. The AI-use certification brings one additional line to pleadings (compared to what the Order VI Rule 15 is duplicating) and has just as much disciplinary power as it pushes verification before signing rather than after challenge. The cost is not too high, the alternative is to keep on repeating episodes of Gummadi Usha Rani and deal with them one after the other, without any institutional learning between them.
- Conclusion
There’s no reason why the signature should be the basis for liability. However, a building can’t be a foundation. The Supreme Court itself has framed AI hallucinations as a matter of misconduct, not error, as a doctrine restoration to remind everyone: When you sign your name to a fabrication, it’s serious. It’s a preliminary change. The next one must be one that extends all the way up to and at the supervisory level and up to the vendor and the advocate at the lectern. The kind of regime (advocate only or tiered) that India will have will influence legal practice for the next generation. The contention for the tiered approach does not that the signature should be less meaningful, but that a new tiered institutional structure must be put in place to keep up with the tools that advocates and judges now rely on every day.
THIS ARTICLE IS WRITTEN BY AYUSH TALWAR FROM OP Jindal Global University
REFERENCE :
Gummadi Usha Rani v Sure Malikarjuna Rao, CRP No. 2487/2025 (AP HC, decided 21 January 2026).
Gummadi Usha Rani v Sure Malikarjuna Rao, 2026 SCC OnLine SC 341(per Narasimha and Aradhe JJ).
Ibid.
Order of the Bombay High Court imposing costs of Rs.50,000 on a litigant for filing AI-hallucinated submissions (January 2026), reported in LiveLaw and MediaNama; Order of the Delhi High Court dismissing a writ petition where opposing counsel exposed fabricated citations (September 2025); Income Tax Appellate Tribunal (Bengaluru), order recalled in December 2024 after discovery of non-existent judgments traced to unverified ChatGPT use.
Ministry of Electronics and Information Technology, Government of India, ‘India’s AI Governance Guidelines’ (February 2026).
Bar Council of India, Standards of Professional Conduct and Etiquette, Part VI, Chapter II, Rules I–IV (framed under s 49(1)(c) of the Advocates Act 1961).
Code of Civil Procedure 1908, Order VI, Rule 15.
Hikmat Ali Khan v. Ishwar Prasad Arya (1997) 5 SCC 201.
R. Muthukrishnan v. Registrar General, Madras High Court (2019) 16 SCC 407.
Advocates Act 1961, s 35.
ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512 (29 July 2024).
American Bar Association, Model Rules of Professional Conduct, Rules 5.1 and 5.3.
Advocates Act 1961, s 49(1)(c).
Consumer Protection Act 2019, ss 2(11) (deficiency of service) and 2(47) (unfair trade practice); Sale of Goods Act 1930, ss 14–16.
Hedley Byrne & Co Ltd v Heller & Partners Ltd [1964] AC 465 (HL). Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules 2026; Digital Personal Data Protection Act 2023
