Key Features of India’s Guidelines for the Use of AI in Patent Examination

Introduction

The Government of India has been issuing various frameworks and guidelines governing the responsible use of Artificial Intelligence (“AI”)[i] in public functions. These frameworks recognise the potential of AI to enhance administrative efficiency and public service delivery, while emphasising human oversight, accountability, safety, transparency, privacy, risk-based governance, and institutional safeguards.[ii]

The Office of the Controller General of Patents, Designs and Trade Marks (“CGPDTM”) issued the Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures[iii] earlier in August 2026 (“Guidelines”) to provide a framework for the responsible use of AI tools by Examiners and Controllers during patent examination. This article informs patent applicants and practitioners about the essential points of the Guidelines, for their better awareness of permitted and responsible use of AI by the patent examiners and controllers, during patent examination processes.

AI-Assisted Output

The Guidelines do not seek to exclude AI from the examination process, nor do they permit its unmonitored use. Rather, they identify where AI may assist, that is, in screening, classification, search, translation support, drafting support, technical comparison, and knowledge retrieval; while making clear that the Examiner’s/Controller’s independent application of mind, confidentiality obligations, and statutory responsibilities cannot be delegated to a machine-based system. The Guidelines are hence drafted with the central philosophy of having a human-in-the-loop approach, i.e., “AI-Assisted Output”[iv].

Limitations, Risks and Human Intervention

The Guidelines identify various limitations and risks relating to the use of AI, which are mentioned below:

Profile Sensitivity and Prompt Discipline:

AI-generated outputs may be influenced by prior interactions, user-specific context and the manner in which a prompt is framed. Outputs may therefore vary depending on the user and the wording, scope and context of the prompt.

Hallucination, Misguidance and Omission:

AI may generate incorrect or fabricated facts, citations, judgements or prior art, or may misinterpret the source material provided to it. It may also omit important claim limitations, amendments, qualifiers or technical relationships, or provide a plausible but incorrect line of reasoning.

False Pattern Matching and Loss of Technical Context:

AI may rely on surface-level similarities while overlooking material legal or technical differences. It may also lose or alter the precise meaning of technical material while summarising or translating it, particularly where specialised patent language, drawings, flowcharts or other technical material is involved.

Errors on Classification, Search Terms and Interpretation Risks:

AI-assisted searches may result in classifications or search terms which are too broad or narrow, resulting in irrelevant prior art or overlooking relevant material.

AI may also have difficulty with complex patent concepts such as claim construction, enablement and inventive step, including jurisdiction-specific legal and procedural nuances.

Black-box Opacity, Data Limitations and Bias:

The basis for an AI-generated output may not always be apparent to the examiner, making verification difficult.

AI tools may also rely on incomplete or outdated datasets and may contain biases arising from their training data, including in relation to particular jurisdictions, technologies, languages or sources.

Confidentiality Breach Risk:

Use of external AI tools may create confidentiality concerns, particularly with respect to unpublished material, internal documents or other confidential information relating to a patentable subject.

The Guidelines emphasise that AI is intended to assist, and not replace, the Examiner or Controller, and suggests as follows:

Any AI-assisted output which may influence search or examination should be critically reviewed and independently validated by the concerned officer, i.e., an emphasis is laid on human oversight and manual verification.
All prompts should be clear, precise and limited to the intended purpose, and the resulting output should be independently assessed by a competent patent examiner/ officer.
Any source or citation generated through AI should therefore be independently verified against the original or other reliable sources, particularly in the case of judicial precedents.
The concerned officer remains responsible for every official act or decision involving the use of AI, and such responsibility cannot be diluted or transferred to the AI tool.

Typical Uses of Artificial Intelligence during Patent Examinations

The Guidelines have consolidated a non-exhaustive list of typical AI use cases together with their potential benefits, risks and corresponding safeguards in the process of patent examination. The same are:

For Classifications, Search Terms, Concept Clusters, and Prior-Art Searches: AI may assist in identifying relevant IPC[v]/ CPC[vi] classification codes, generating synonyms, related expressions and technical terminology, and conducting preliminary prior-art searches through officially subscribed patent databases. However, these must be assessed by the concerned Examiner for relevance and verified against authenticated classification sources, as well as manually checked against the claims and complete specification. AI may also support prior-art searches through AI-backed databases as subscribed by the Patent Office. The Examiner shall remain responsible for such searches and selections.

For Claim Analysis, Feature Extraction and Preliminary Novelty/Inventive-Step Assessment: AI may assist in breaking a claim into individual features, preparing preliminary claim mappings, and organising a structured comparison for novelty or inventive-step analysis. However, it may split the claim language incorrectly or substantially alter the technical nuance.

Therefore, it is necessary that the identified features and mappings are manually verified against the original claim and relevant documents. No conclusion on novelty or inventive-step analysis shall be based solely on unverified AI-generated output, and the final assessment shall remain that of the Examiner.

For Translation and Technical Understanding: AI may be used for a preliminary understanding of foreign-language or technically complex documents.

Where an AI-generated translation is relied upon for a formal objection or examination purpose, it must be verified manually against the original document, drawings or other reliable material. The use of machine translation, where relied upon, should be recorded in patentability’s reasoning, particularly where mistranslation or loss of technical nuance may affect the interpretation.

For Drafting, Language and Clarity Review: Once an officer has prepared the substantive draft and reasoning, AI may be used to improve grammar, coherence, and formatting, and to identify potential ambiguity or drafting issues in a claim. AI must not introduce or alter the underlying legal or technical reasoning.

Any issue identified by AI must be independently assessed by the Examiner before being relied upon or raised as an objection, and the final draft must be thoroughly reviewed before issuance.

For Legal, Technical and Patent Citations and Case Laws: AI may assist in identifying possible case laws, scientific references, patent citations, manuals and related legal or technical concepts, as well as in obtaining a preliminary understanding of a judgement. However, it must be ensured that no citation, quotation, ratio or legal proposition generated or identified may be used unless independently verified against the original source.

Judgments must also be considered in the context of their facts and circumstances, and the Examiner must independently determine the relevance and applicability of any precedent. Any fabricated or unverifiable source or proposition must be discarded.

Re. Confidentiality and Use of Unpublished Patent Material: Unpublished patent applications and other confidential material must not be entered into Public AI Tools[vii]. Any approved internal or secured AI tools may be used for such material.

It shall be the responsibility of the Examiner or officer to ensure that no confidential or unpublished material is disclosed to any unapproved external system and, where there is a doubt, such material should not be entered into the AI tool.

For Preliminary Assessment of Sufficiency of Disclosure: Assistance from AI may be sought in identifying possible gaps in enablement, fair basis or other aspects of sufficiency of disclosure. However, the complete specification, drawings, and relevant common general knowledge must nonetheless be examined independently.

Further, the Examiner must independently assess whether the alleged gap is real, material, and relevant to the statutory requirement on legal and technical basis before any objection is framed.

Prohibited Uses of AI

The Guidelines particularly prohibit the following, during patent examinations:

Entering unpublished/ confidential material relating to an invention into public AI tools.

Substituting AI based assessment for the officer’s independent application of mind on substantive matters, including novelty, inventive step, industrial applicability, sufficiency, etc.

Issuing office actions, decisions and other official communications on the basis of AI-Assisted output alone.

Relying on unverified citations, including prior arts and caselaw.

Using AI-generated content without review.

Relying solely on AI for decision-making and affecting the rights of parties, especially where third-party rights (such as patent oppositions) are involved.

Administration Of Artificial Intelligence Tools

To operationalise the above principles, the Guidelines provide for certain mechanisms.

The Guidelines state that the competent authority may prescribe that the material use of AI in specific functionalities should be recorded. The records may include the specific AI tools utilised, nature of usage, date of usage, and the like, that may be deemed to be necessary for supervision, audit and quality review. Pertinent to mention that such disclosure may be made available to stakeholders to improve confidence in the examination process.

For the governance of the use of AI in the Patent Office during examinations, the Guidelines provide for the institution of dedicated AI Governance Committee (with specific roles and responsibilities[viii]), comprising of examiners and controllers from the Examination Division..

In order to effectively enforce human oversight over AI-Assisted Output, the authority may provide for the training of patent officers on different aspects of AI uses and its prohibitions, including regarding the functionalities and limitations of AI, confidentiality risks and prompt discipline, verification of AI-assisted outputs, and the identification of hallucinations and fabricated citations.

Further, where appropriate, the competent authority may provide for independent audit, impact assessment, feedback mechanisms, and incident-reporting systems.

Concluding Remarks

The Guidelines reflect an attempt to reconcile efficiency and quality of patent examination with accountability, while also attempting to keep up with the evolving legal and technological landscape. In the present scenario, it is unwise to not make use of such AI tools to one’s advantage. The key takeaway from the Guidelines is that AI should be used as an assistive tool and not become the decision maker.

Having said that, the Guidelines essentially seek a balance between technological efficiency, legal duties, and the rights of innovators, patent applicants, patentees and other affected parties.

The effectiveness of the Guidelines will largely depend not only on the capabilities of the AI tools adopted, but on how effectively the safeguards, verification requirements and accountability mechanisms are implemented in practice.

Contributed by: Akash Sajan, Shubhra Satpathy, and Geia Natraj, with guidance from Aditi Verma Thakur.

[i] “Artificial Intelligence” means a machine-based system capable of generating outputs such as predictions, recommendations, classifications, summaries, or content on the basis of input data along with prompts.

[ii] NITI Aayog, Government of India, RESPONSIBLE AI #AIFORALL, Approach Document for India: Part-1 & Part-2.

[iii] Office of the Controller General of Patents, Designs and Trademarks, Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures. August 7^th, 2026.

[iv] “AI-assisted output” refers to content, decisions or products generated through a collaborative process where AI tools are used to enhance, speed up or augment human work. Unlike fully automated output generation by AI tools, AI-assisted output involves “human in the loop”, meaning a person provides direction, context, and oversight, ensuring the final result meets specific quality and intended standards.

[v] IPC means International Patent Classification.

[vi] CPC means Cooperative Patent Classification.

[vii] “Public AI tool” means an AI tool or application that is accessible to the public through free or commercially available means, including both free-tier and paid versions of consumer-facing AI platforms. Such tools are trained on large and diverse datasets from publicly available sources, including text, images, audio, video, and other digital content. “Private AI tools” refer to artificial intelligence systems operating in closed or restricted environments, often trained on proprietary or controlled datasets. Such systems are typically owned and managed by specific entities who control the data feeding the system and retain the IP of the models involved.

[viii] The roles and responsibilities of the committee include review and approval of AI tools, categorisation of permitted and prohibited uses of AI, prescription of safeguards, conducting pilot studies, review of quality issues and complaints, periodical revision of the Guidelines, and consultation with AI experts.