The publication of the Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures by the Office of the Controller General of Patents, Designs and Trade Marks on 7 August 2026 marks the first occasion on which the Indian Patent Office has formally acknowledged and attempted to govern, a practice that had already been quietly taking hold within the examination process. Examiners and Controllers across the patent offices in Delhi, Mumbai, Kolkata and Chennai had, in the ordinary course of an increasingly AI-saturated professional environment, begun reaching for generative AI tools to assist with searching, classifying, translating, drafting and reasoning through applications a development the Patent Office had neither sanctioned nor regulated until this sixteen-page document, accompanied by illustrative annexures and an officer checklist, was issued. For a quasi-judicial authority whose findings on novelty, inventive step and sufficiency of disclosure carry direct and often substantial commercial consequences for applicants, the absence of any prior governance framework around AI use was a regulatory gap of considerable significance and the Guidelines represent the Patent Office’s first sustained attempt to close it.
The stakes attached to getting this framework right are not merely administrative. Patent examination under the Patents Act, 1970 is a statutory, quasi-judicial function the Controller’s findings on anticipation under Section 2(1)(j) and Section 2(1)(l), on inventive step under Section 2(1)(ja) and on sufficiency of disclosure under Section 10(4), culminate in orders under Section 15 or Section 43 that determine whether an applicant secures a twenty-year monopoly or is denied one. The introduction of AI assistance into this process, however carefully circumscribed, raises genuine questions about the integrity of the examiner’s independent application of mind, the transparency of the reasoning an applicant is entitled to receive and the adequacy of existing appellate and revocation mechanisms to address errors whose origin may be difficult to trace to a human or a machine. This article examines the 2026 Guidelines in detail their scope, their permitted and prohibited uses of AI, their governance architecture, their interaction with the pre-existing Revised Guidelines for Examination of Computer Related Inventions, 2025 and the practical and institutional concerns the Guidelines have already generated among Indian patent practitioners before assessing their likely impact on the conduct of patent prosecution in India going forward.
Distinguishing the AI Guidelines from the CRI Guidelines
A threshold clarification is necessary because the two documents are easily and mistakenly, conflated. The Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures, issued in August 2026, address the procedural question of how Examiners and Controllers may use AI tools as an aid in carrying out examination functions. They are entirely distinct from the Revised Guidelines for Examination of Computer Related Inventions, which the CGPDTM released in 2025 and which address the substantive question of when an invention involving artificial intelligence, machine learning or blockchain technology satisfies the patentability requirements of Section 3(k) of the Patents Act, 1970 the statutory exclusion for mathematical methods, business methods, algorithms and computer programmes per se. One document governs the examiner’s tools; the other governs the patentability of the applicant’s invention. Practitioners handling AI-related patent applications must keep this distinction sharply in view, since the two sets of guidelines apply at entirely different stages and to entirely different questions within the same examination file.
The Twelve Permitted Uses – Four Functional Categories
The 2026 Guidelines identify twelve distinct uses of AI across the examination lifecycle, organised in practitioner commentary into four functional categories that track the structure of the examination process itself. The first category, patent search and prior-art identification, permits the use of AI to identify candidate International Patent Classification and Cooperative Patent Classification codes, to generate search terms and concept clusters for a given invention and to conduct prior-art searches using officially subscribed AI-enabled search platforms. The second category, claim and patentability analysis, is considerably more consequential and permits preliminary claim-feature extraction, preliminary novelty or inventive-step analysis, quick identification of clarity issues in claims and preliminary assessment of the sufficiency of disclosure under Section 10(4). The third category, drafting, language and legal research support, permits translation support for non-English prior art, assistance in improving the structure or language of a draft Office communication, generation of legal or technical citations and assistance in understanding case law and related legal or technical concepts. The fourth category is, in substance, a confidentiality constraint rather than a permitted use it establishes an absolute prohibition against the use of public generative AI tools in connection with unpublished patent application material.
The Guidelines draw, whether explicitly or through the structure of their caveats, a meaningful distinction between the first and third categories on the one hand and the second category on the other. The uses grouped under patent search and under drafting and legal research support are predominantly preparatory and organisational they reduce the time an officer spends on tasks that are largely mechanical in character and do not themselves constitute the substantive evaluative judgment that patent examination requires. The uses grouped under claim and patentability analysis are qualitatively different, because preliminary novelty and inventive-step analysis and the preliminary assessment of sufficiency of disclosure, sit at the very core of the Controller’s quasi-judicial function under the Patents Act, 1970. It is here that the Guidelines’ permission for AI assistance has attracted the most sustained professional scrutiny.
Novelty, Inventive Step and the Problem of Cognitive Anchoring
For novelty and inventive-step analysis, the Guidelines permit AI platforms to assist with preliminary claim mapping and structured comparison against identified prior art, while requiring that each mapped feature be independently verified by the officer against the actual prior-art document and that the final determination on novelty or inventive step rest on the officer’s own analysis and application of mind. The Guidelines themselves acknowledge the specific risks this use carries misreading of cited documents and hindsight bias chief among them and require independent verification as the primary safeguard.
The structural difficulty this raises, which Indian IP commentators have identified with some precision, is that verification occurring after an officer has already been presented with an AI-generated analytical frame is not equivalent to an independent assessment conducted without that frame. Once an AI tool has proposed that a particular claimed feature corresponds to a particular prior-art disclosure, the officer’s subsequent engagement with the prior-art document risks becoming an inquiry into whether that proposed correspondence can be sustained, rather than an open inquiry into whether the reference, read as a whole and on its own terms, in fact discloses the claimed feature. This concern is sharper still for the inventive-step inquiry than for the novelty inquiry, because inventive step under Section 2(1)(ja) of the Patents Act, 1970 requires an assessment of the invention as a whole, judged through the eyes of a person skilled in the art, without the benefit of hindsight a standard that Indian courts have repeatedly emphasised cannot be satisfied by reconstructing the invention from its own disclosure and then searching backward for corroborating prior art. An AI tool built for pattern matching, similarity detection and retrieval across large prior-art corpora is, by its very design, susceptible to exactly this kind of hindsight-driven mosaicing, a risk the Guidelines themselves flag without resolving through any structural safeguard beyond the instruction to verify. A comparable concern attaches to the permitted use of AI in assessing sufficiency of disclosure under Section 10(4): identifying a gap in a specification’s enablement or fair basis is not the same exercise as determining whether that gap is legally material and an AI tool’s flagging of a perceived enablement gap risks framing the officer’s inquiry from a deficit-oriented starting point rather than from an independent assessment of the specification as a whole, considered alongside the common general knowledge of the person skilled in the art.
Confidentiality, Public and Private AI Tools and Data Governance
The Guidelines draw a sharp distinction between public generative AI tools openly accessible consumer-facing platforms and private AI tools, which the Guidelines describe as closed, proprietary systems, including AI features embedded within the commercial patent search databases to which the Patent Office subscribes. The Guidelines treat private tools as inherently more suitable for official use on account of their tighter data security and their more contextually attuned handling of patent-specific prompts and they impose an absolute prohibition on entering unpublished patent application content, trade secrets or internal deliberative material into public generative AI tools a safeguard directed at the obvious risk that confidential, commercially sensitive and pre-publication technical disclosures could otherwise leak into training data or be exposed through a public platform’s logging and retention practices.
This confidentiality architecture, while sound in principle, has attracted criticism on transparency grounds. The Guidelines decline to name the specific private AI platforms accessible to the Patent Office, stating that such tools are proprietary in nature and have accordingly not been specifically identified a position Indian commentators have characterised as conflating the separate questions of a tool’s proprietary status and the public disclosure of its identity, since a commercial platform can be both proprietary and publicly named, as is routine practice for government procurement of commercial software elsewhere. The Guidelines are similarly silent on the procurement process through which these private AI tools were acquired, on data retention and processing arrangements, on whether prompts and outputs submitted by officers are used by third-party providers to improve or train their own models and on what contractual data-protection restrictions, if any, the Patent Office has imposed on such providers questions of genuine data-governance significance given that at least one commercial patent database subscription held by the Office, for the Deterrent World Patents Index and DesignVision databases, is provided by a foreign analytics company, raising tech-sovereignty considerations that extend beyond the immediate question of examiner conduct.
Hallucination, Fabrication and the Guidelines’ Own Drafting
The Guidelines devote specific attention to the risk of AI hallucination and fabrication the tendency of generative AI tools to present confident but unsupported statements, citations, explanations or claim-to-prior-art mappings. The Guidelines attempt to distinguish extrinsic hallucination, where the AI tool fabricates information not present in the context it was given, from intrinsic hallucination, where the AI tool is given specific source documents but its output distorts or misrepresents what those documents actually say a distinction of genuine practical importance to examiners cross-checking AI-generated legal or technical citations, prior-art summaries and claim-feature mappings against the underlying source material, though Indian IP commentators have noted that the Guidelines’ own drafting of this distinction contains an internal inconsistency in its use of the terms “intrinsic” and “extrinsic.” For practitioners, the operative point remains sound regardless of this drafting slip: AI-generated outputs in a patent examination context may be confidently wrong in two distinct ways and verification against source material is the Guidelines’ prescribed, if imperfect, response to both.
Administrative Measures, the AI Governance Committee and the Accountability Gap
The Guidelines establish an AI Governance Committee within the Patent Office, comprising examiners and Controllers from the Examination Division, officers from the IT Office and officers from the Quality Management System Division, charged with the institutional oversight of AI use across the Office. The Guidelines also impose a record-keeping obligation on individual officers: whenever AI is used in a specified examination function, the officer is required to record the name and nature of the tool used, the date of use and such other information as may be necessary for supervision, audit or quality review. The Patent Office is further required to securely retain the prompts, inputs, outputs, tool details, prior-art results, generated drafts and the officer’s own treatment of those materials whenever AI materially influences examination and is expected to disclose such material contribution when it issues the document containing the AI-assisted work.
This administrative architecture, while establishing a genuine audit trail, leaves open a significant accountability question that Indian practitioners have already identified. The Guidelines do not require disclosure of AI use within the body of a First Examination Report or a final order itself, meaning an applicant reading an adverse examination report has no way of knowing, from the face of the document, whether a particular novelty objection or inventive-step finding was shaped by an AI-generated claim mapping. Nor do the Guidelines specify any corrective or remedial mechanism to be invoked where a material AI-related error a hallucinated citation, a misleading prior-art mapping that unduly influenced an officer’s reasoning comes to light, beyond the ordinary statutory channels of response to a First Examination Report, pre-grant or post-grant opposition under Section 25 and appeal. Where an applicant suspects that an adverse finding rests on an AI-generated error embedded invisibly within an officer’s reasoning, the practical path to establishing and remedying that error through the existing statutory framework is far from straightforward a concern that echoes and in some respects compounds, the pre-existing and well-documented problem of inadequately reasoned patent grant and rejection orders issuing from the Indian Patent Office, a problem to which the introduction of an unseen AI-generated analytical layer adds a further dimension of opacity.
Impact on Patent Prosecution Practice in India
For practitioners prosecuting applications before the Indian Patent Office, the practical consequences of the 2026 Guidelines are immediate and warrant specific adjustments to prosecution strategy. First, responses to First Examination Reports containing novelty or inventive-step objections should proceed on the assumption that an AI-generated preliminary claim mapping may underlie the Controller’s stated reasoning, even where the report itself gives no indication of AI involvement and practitioners should accordingly scrutinise cited prior art with particular care for mischaracterisation of disclosed features, mosaicing of multiple references without the hindsight-free justification Indian jurisprudence requires and citations that, on inspection, do not actually support the proposition for which they are invoked. Second, where a FER’s treatment of sufficiency of disclosure appears to rest on an asserted enablement gap that does not withstand scrutiny against the complete specification, drawings and the common general knowledge of the person skilled in the art, practitioners have a stronger basis than before to challenge the characterisation of that gap as material, given the Guidelines’ own acknowledgment that AI-flagged gaps require independent technical and legal assessment before they can properly ground a sufficiency objection. Third, practitioners should factor the record-keeping obligations into any strategy involving a Right to Information request directed at understanding the basis of a particular examination outcome, since the Guidelines now create an identifiable category of retained material prompts, AI outputs and the officer’s treatment of them that did not previously exist as a documented record. Fourth and more broadly, the Guidelines’ emphasis on private, officially sanctioned AI tools over public generative AI platforms signals that the Patent Office’s own use of AI will likely expand rather than contract over the coming examination cycles, making early and informed engagement with the substance of AI-assisted objections, rather than a generic objection to AI’s involvement as such, the more productive posture for applicants and their counsel.
Conclusion
The 2026 Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures represent a genuinely significant, if incomplete, step in bringing formal governance to a practice that had already begun operating within the Patent Office without any public framework. Their stated foundational principle that AI may assist Examiners and Controllers but cannot substitute for the independent application of mind that the Patents Act, 1970 requires of a quasi-judicial examining authority is sound and the Guidelines’ express prohibitions on issuing First Examination Reports, hearing notices or final decisions based solely on unvalidated AI output and on exposing unpublished application material to public AI tools, address the most acute risks directly. What the Guidelines have not yet resolved and what Indian patent practitioners, academics and the Patent Office itself will need to continue working through in the examination cycles that follow this framework’s introduction, is the harder question of institutional accountability once human oversight fails in a given case: whether the record-keeping obligations the Guidelines impose translate into a meaningful corrective mechanism for the applicant whose application was shaped by an AI-generated error and whether the existing statutory architecture of opposition, revocation and appeal under the Patents Act, 1970 is adequate to a world in which the precise origin of an examiner’s reasoning human, machine-assisted or some increasingly difficult-to-disentangle combination of the two may no longer be straightforwardly ascertainable from the face of the examination record. The trajectory of Indian patent practice over the next several years will be shaped substantially by how these questions are answered, whether through refinement of the Guidelines themselves, through judicial scrutiny of AI-influenced examination outcomes as they begin to reach the High Courts or through further institutional reform of the Patent Office’s internal review and quality-assurance mechanisms.
References
- Patents Act, 1970 – https://ipindia.gov.in/pages/patents/chapter
- Patents Rules, 2003 (as amended) – https://ipindia.gov.in/acts/patent-rules-2003
- Office of the Controller General of Patents, Designs and Trade Marks Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures (7 August 2026) – https://ipindia.gov.in/storage/uploads/docs-operator/ade11dcd-c118-41f4-ab58-cc150f47a736.pdf
- Office of the Controller General of Patents, Designs and Trade Marks Revised Guidelines for Examination of Computer Related Inventions, 2025 – https://ipindia.gov.in
- Intellectual Property India Official Portal – https://ipindia.gov.in
- Delhi High Court Judgments Portal – https://delhihighcourt.nic.in
- Department for Promotion of Industry and Internal Trade National IPR Policy 2016 – https://www.dpiit.gov.in/ministry/about-us/details/Title=National-Intellectual-Property-Rights-(IPR)-Policy-ITMwETMtQWa
- World Intellectual Property Organization – Patents – https://www.wipo.int/patents/en/
- Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) – https://www.wto.org/english/docs_e/legal_e/27-trips.pdf
Compulsory Licensing Copyright Act Copyright Act 1957 copyright case laws Copyright Enforcement Copyright Infringement Copyright Law Copyright Rules Descriptive Marks India Indian IP Framework Indian Patent Act Indian Patent Law Indian Trademark Law Intellectual Property Law IP Law India Patent Claims Patent Enforcement Patent Infringement Patent law Patent Revocation Patents Act Patents Act 1970 Pharmaceutical Patents Section 9 Trade Marks Act Section 29 The Patent Act 1970 Trademark Infringement Trademark Registration Trade Marks Act 1999 TRIPS Compliance