Summary
The Indian Intellectual Property Office published the Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures on 7 August 2026. The Guidelines relate to the use of Artificial Intelligence by Examiners and Controllers while carrying out different functions during patent examination. They deal with the circumstances in which AI may assist, the risks that may arise from its use, the safeguards contemplated by the Indian IP Office, the uses that are prohibited, and the responsibility that continues to remain with the Examiner or Controller. An outline of what the Guidelines cover is provided in this article.
Introduction to the Guidelines
The Indian Intellectual Property Office published the Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures on 7 August 2026. The Guidelines relate to the use of Artificial Intelligence in the patent examination process and describe how Examiners and Controllers may use AI for different examination related functions.
The Guidelines cover several stages of patent examination. They address classification, searching, claim feature extraction, translation, drafting support, preliminary novelty and inventive step analysis, clarity, sufficiency of disclosure, citation retrieval, and understanding legal and technical concepts. They also set out the risks identified by the Indian IP Office, the safeguards to be followed, the uses that are prohibited, and the responsibility of officers using AI.
An outline of these aspects of the Guidelines is provided below.
AI and Patent Examination
The Guidelines describe patent examination as document intensive, technically complex, legally evolving, and time sensitive. According to them, AI tools may assist in suggesting search terms, retrieving and ranking relevant documents, visualising content, and reducing the time and cost involved in screening large numbers of patent documents. They also refer to recent work considering the use of AI in novelty analysis.
The Guidelines state that they have been prepared in the background of growing use of generative AI and Government policy supporting responsible use of Artificial Intelligence in public functions. Their stated purpose is to identify how the benefits of AI tools may be appropriately used in patent examination without compromising the technical, statutory, and quasi judicial functions of the Examiner and Controller.
Their stated objective is to guide and regulate the use of Artificial Intelligence in patent examination in a manner that supports efficiency and quality while preserving confidentiality, accountability, consistency, and independent application of mind.
The scope extends to screening, classification, search, translation support, drafting support, technical comparison, and knowledge retrieval.
The Guidelines also distinguish between public and private AI tools. Public AI tools include consumer facing applications available freely or commercially. Private AI tools operate in closed or restricted environments and may use proprietary or controlled data. The Guidelines state that private systems, because of their restricted environment and security controls, may be better suited for processing sensitive material.
That distinction becomes particularly relevant to patent examination because the Patent Office handles unpublished patent applications, internal notes, and other material that may not be available to the public.
Where AI May Assist an Examiner
The Guidelines identify several examination functions in which AI may provide assistance.
For patent classification, they state that AI may generate candidate IPC or CPC classifications. At the same time, they record that classification outputs may differ between models and may suggest erroneous, irrelevant, or overly broad classes. The classification is therefore to be checked against an authenticated source and manually verified with the claims read together with the complete specification.
AI may also assist in generating search terms and concept clusters. According to the Guidelines, an AI tool may suggest synonyms, related expressions, alternative spellings, broader terms, and connected technical concepts. Such suggestions may help in developing search vocabulary, particularly because terminology differs across patent documents, scientific literature, standards, and jurisdictions.
The Guidelines nevertheless place selection of the final search terms with the Examiner. They state that the officer must decide which terms remain sufficiently close to the claimed subject matter, which expressions are merely technically adjacent, which broader terms may assist exploratory searching, and which terms may divert the search towards irrelevant prior art.
Claim feature extraction is another contemplated use. AI may break a long claim into individual features for comparison or internal organisation. The Guidelines point out that an AI system may incorrectly split claim language, overlook relationships between elements, miss numerical ranges or negative limitations, or separate features that have to be read together. They accordingly state that novelty or inventive step analysis must not proceed on an unverified AI generated feature list.
Translation support is also contemplated. An Examiner or Controller may use AI to obtain a rough understanding of a foreign language document. The Guidelines state that technical meaning may be lost in machine translation and that any translated material relied upon for patentability has to be considered with the surrounding disclosure and drawings where necessary. They also state that reliance on machine translation should be recorded in the reasoning on patentability.
For office communications, the Guidelines permit AI assistance for grammar, coherence, formatting, headings, readability, and concise expression after the officer has prepared the substantive draft or reasoning. They caution that AI may insert unsupported statements, alter the officer’s reasoning, use legally imprecise language, or generate nonexistent citations.
The Guidelines also contemplate AI assistance for identifying possible legal or technical citations, preliminary novelty or inventive step analysis, identification of clarity issues, prior art searching through officially subscribed AI tools, preliminary assessment of sufficiency of disclosure, and understanding case law and technical or legal concepts.
Across these uses, the Guidelines treat AI output as an input to the examination process and place the final assessment with the Examiner or Controller.
Risks of AI and Human Judgment
A substantial part of the Guidelines deals with the manner in which AI output may go wrong.
One of the risks identified is hallucination or fabrication. According to the Guidelines, an AI tool may present an unsupported statement, citation, explanation, or mapping with confidence. In patent work, this may include nonexistent prior art, fabricated judgments, incorrect claim mappings, or misstatements about technical disclosures.
The Guidelines state that sources cited by AI have to be checked against authorised or original sources. They pay particular attention to judicial precedents because an AI system may misread the ratio of a judgment, overstate a legal proposition, or apply it outside the factual setting in which the court stated it.
Other risks identified by the Guidelines include omission of a critical claim feature, false pattern matching, loss of technical context, classification and search drift, over broad or narrow interpretation, black box opacity, confidentiality problems, misguidance, incomplete or outdated data, bias, and difficulty in understanding specialised patent language.
The Guidelines also discuss prompt sensitivity. The manner in which a question is framed may materially influence the output.
They illustrate this point through a novelty exercise involving a laparoscopic needle holder. In the example, one prompt led the AI system to conclude that the claim lacked novelty over a prior art document. A later prompt specifically directed attention to whether the prior art disclosed a detachable attachment between the proximal and distal shaft portions. The AI then concluded that the claim was novel because that feature was not disclosed.
The Guidelines state that both responses appeared fluent and persuasive, but differed in their treatment of the detachable attachment feature. They use the example to state that AI responses may not always be consistent or reproducible and that final patent analysis requires human judgment.
The same approach appears in the general provision on human oversight. The Guidelines state that Artificial Intelligence is intended to assist and not replace the functions or work of the Examiner or Controller. Any AI use that may influence search or examination is to undergo manual review and validation, and the Examiner or Controller remains responsible for official acts involving AI.
Confidentiality, Verification and Prohibited Uses
The Guidelines place clear restrictions on the use of public AI tools with unpublished patent material.
They state that unpublished patent application contents, confidential office records, and internal deliberative material must not be entered into public AI tools. They separately explain that an officer should not copy claims, descriptions, prior art notes, hearing notes, or unpublished documents into a public generative AI tool for obtaining a summary, chart, or draft.
According to the Guidelines, doing so creates a risk that unpublished subject matter, internal comments, or other sensitive official material may move outside the approved environment. They classify this use as prohibited.
The Guidelines also state that Artificial Intelligence cannot be used as a substitute for the Examiner’s or Controller’s application of mind on substantive matters such as novelty, inventive step, industrial applicability, sufficiency, clarity, or unity of invention.
They state that office actions, First Examination Reports, hearing notices, decisions, and other official communications should not be issued solely on the basis of AI generated output without adequate human oversight. AI generated content used in an official communication has to be reviewed, corrected, and adopted by the concerned officer.
The Guidelines extend the restriction to decisions affecting applicants, patentees, and third parties. They specifically refer to situations involving opposing parties, such as patent opposition proceedings, and state that decisions affecting rights should not rest solely on Artificial Intelligence.
Verification of citations receives particular attention. The Guidelines state that case law, prior art, scientific literature, or other references suggested by AI should not be used unless independently verified from authentic sources. Where verification is not feasible, they state that the AI generated citation should be discarded.
The Guidelines illustrate this problem with an experiment involving a patent judgment. A generative AI tool was asked to provide verbatim quotations relating to the concept of teaching away. It initially presented passages as verbatim extracts and attributed them to a specific judgment. When the user stated that the passages could not be found, the tool eventually acknowledged that the passages were reconstructions rather than direct extracts from the decision.
The Guidelines describe this as an example of fabricated verbatim quotations and state that AI generated quotations should not be trusted without verification against the primary source.
Practical Examples and Officer Accountability
The Guidelines contain a series of practical exercises involving actual patent examination functions rather than limiting themselves to general principles.
One exercise compares classifications generated by different AI tools and prompt configurations for the same patent applications. The classifications varied between systems. The Guidelines record that the suggestions were model sensitive both in the number of classification symbols generated and in the technical direction taken by the models. They also state that AI suggested classifications were sometimes close to those assigned manually, but were not identical.
Based on those examples, the Guidelines state that human judgment continues to remain necessary because final classification involves reading the claims and specification and identifying the appropriate technical focus of the invention.
A similar exercise deals with search terminology. Different generative AI systems were asked to produce search terms for the same claim involving arsenic removal using activated carbon supporting magnetic iron oxide nanoparticles. Although the systems identified the main technical concepts, their suggestions differed in breadth, terminology, and movement into adjacent technical areas.
The Guidelines state that the Examiner has to determine which terms stay sufficiently close to the claimed subject matter and which may introduce irrelevant prior art or divert the search.
The Guidelines also give examples relating to novelty, sufficiency of disclosure, clarity objections, translation, and retrieval of case law. These examples are used to show possible benefits of AI assistance as well as errors that may arise if output is accepted without verification.
Annexure II takes the safeguards into the examination workflow through a checklist and declaration.
The checklist asks whether the proposed use is permitted under the office guidelines, whether confidential material has been excluded, whether AI is being used only for assistance, whether technical, factual, and legal statements have been verified, whether cited sources have been checked, and whether the final conclusion remains the result of the officer’s independent application of mind.
The proposed module also asks an officer whether Artificial Intelligence was used in a particular work product and the purpose for which it was used. The listed uses include classification, search support, technical support for preliminary understanding, language refinement, and other permitted assistive uses.
The accompanying declaration states that the officer remains responsible for the search, citations, analysis, reasoning, conclusions, and official text and that these have been independently reviewed and adopted on the basis of the officer’s own application of mind.
AI Governance in the Patent Office
The Guidelines contemplate institutional measures for managing AI use within the Patent Office.
They state that the competent authority may require material AI use in specified functions to be recorded. Such a record may contain the name of the tool, nature of use, date of use, and other information considered necessary for supervision, audit, or quality review. The Guidelines also contemplate making such disclosure available to stakeholders to improve confidence in the examination process.
They further provide for the possibility of an AI Governance Committee consisting of Examiners and Controllers from the Examination Division, officers from the IT Office, and officers from the Quality Management System Division, with representation covering different technology groups.
Functions contemplated for the committee include reviewing and approving AI tools, categorising permitted and prohibited uses, prescribing safeguards and approval conditions, undertaking pilot studies, reviewing quality issues and errors, considering complaints or policy breaches, revising the Guidelines periodically, and holding consultations with stakeholders and AI experts.
The Guidelines also contemplate training and capacity building for Patent Office officers. The subjects identified include functionality and limitations of Artificial Intelligence, confidentiality risks, prompt discipline, verification of AI assisted output, responsible use in public service, identification of hallucinations and fabricated citations, and permitted and prohibited uses.
They further state that independent audits, impact assessments, feedback mechanisms, and incident reporting systems may be provided where appropriate.
Public Comments and Stakeholder Participation
The Guidelines published on 7 August 2026 seem not to have been placed open for public comment before publication. Patent examination involves several stakeholders apart from the Patent Office. Applicants and patentees are directly affected by examination. Patent agents and attorneys participate in prosecution and hearings. Foreign Patent Offices may interact with the Indian Patent Office through international search, examination, work sharing, and related cooperation. Researchers, technology companies, public interest groups, competitors, and members of the general public may also have an interest in how AI is used in a process that determines the scope and grant of patent rights.
As the Guidelines address matters such as AI assisted prior art searching, preliminary patentability analysis, machine translation, citation verification, confidentiality, disclosure of AI use, and the manner in which AI may support official communications, stakeholder comments could provide practical inputs from different sides of the patent system.
Making the Guidelines available for public comments could therefore add value by giving the Patent Office the benefit of experience and perspectives from applicants, practitioners, industry, foreign offices, technical experts, researchers, and the public. Such a process could also assist future revisions contemplated by the Guidelines themselves as AI technologies and their use in patent examination continue to develop.
Reference
Office of the Controller General of Patents, Designs and Trade Marks, Government of India, Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures, published 7 August 2026, IP India.
Accessible version of the Guidelines: Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures, Accessible Version may be downloaded here.
Disclaimer
This article is based on the author’s personal understanding of the subject. Others may hold different opinions or understandings. This article is intended for general information and does not constitute legal advice. Readers should consult an attorney before acting on any legal issue. An AI application was used to generate parts of this article based on user inputs and prompts.