A Fundamental Shift in the IP Landscape
Discussions surrounding Artificial Intelligence (AI) in intellectual property usually centre on whether AI-generated inventions can be patented or whether an AI system can be named as an inventor. However, a far more immediate operational shift is occurring within the administrative machinery of the Indian Patent Office. The Office of the Controller General of Patents, Designs and Trade Marks (CGPDTM) recently issued official Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures. This regulatory framework establishes clear boundaries for how Examiners and Controllers may or may not leverage AI tools during prior art searches, classification, claim feature extraction, and draft preparation. For patent applicants, startups, and R&D-driven enterprises, this development directly impacts how applications will be analysed, objected to and prosecuted in India.
Overview of the CGPDTM AI Examination Framework
The Guidelines establish a comprehensive framework for the use of AI across the entire examination lifecycle. Specifically, the guidelines delineate permissible assistive uses, including candidate IPC/CPC classification, keyword and concept-cluster search generation, preliminary claim feature extraction, machine translation of foreign prior art, readability support for draft office communications, preliminary novelty and inventive step screening, and legal citation retrieval, while systematically identifying technical risks such as search drift, hallucinated citations, and distorted claim construction.
Crucially, the guidelines establish strict procedural safeguards and explicit prohibitions: mandating that AI cannot replace independent human application of mind, barring official actions or decisions based solely on unvalidated model output, prohibiting the input of unpublished patent disclosures into public AI tools to protect confidentiality, and introducing administrative oversight mechanisms, such as usage logging and an AI Governance Committee to ensure transparency, data security, and institutional accountability across Indian patent prosecution.
A crucial protection highlighted in the Guidelines is the absolute prohibition against entering unpublished patent application content, trade secrets, or internal deliberative notes into public AI models. Only secure, officially authorised internal environments are permitted for official handling.
The Core Mandate: Human Review Remains Necessary
The foundational principle running through the CGPDTM guidelines is unambiguous: AI tools are intended strictly for administrative and search assistance; they cannot replace independent human judgment. While examiners may use approved AI applications to cluster concept terms, format preliminary feature charts, or generate candidate International Patent Classification (IPC) codes, the legal and technical responsibility for every official act remains entirely with the officer. Controllers and Examiners are expressly prohibited from issuing First Examination Reports (FERs), hearing notices, or final decisions based solely on unvalidated AI outputs.
For applicants, this creates an important procedural safeguard. If an examination report raises a dubious objection or relies on an off-target prior art document with a clear impression of AI-generated text, the applicant has a legitimate right to demand a reasoned, human-led analysis rooted firmly in the provisions of the Patents Act, 1970.
Critical Examination Vulnerabilities to Monitor
While AI tools may help to streamline examination timelines, probabilistic models introduce specific failure modes into the prosecution process that applicants must actively audit:
- Classification & Search Drift: AI tools frequently suggest IPC/CPC categories based on surface-level keyword matching. If an Examiner adopts a flawed AI classification, the prior art search drifts into adjacent, irrelevant fields, retrieving non-analogous art while missing critical primary references.
- Distorted Claim Construction: Patent claims rely on precise structural, functional, and relational phrasing (e.g., “wherein”, “configured to”, “operably connected”). Automated feature-extraction algorithms often strip these contextual ties, leading to oversimplified or incorrect novelty and inventive-step objections.
- Hallucination and Source Errors: Generative models can present inaccurate information with high fluency, misinterpreting judicial precedents, fabricating technical citations, or misquoting foreign patent disclosures.
- Machine Translation Distortions: While AI assists in screening non-English prior art (such as Japanese, Korean, or Chinese disclosures), machine translations frequently collapse subtle legal distinctions between technical terms like “attached”, “bonded” or “coupled”.
Strategic Action Plan for Patent Applicants and Inventors
We strongly recommend that applicants enforce equivalent internal governance. Inventors, R&D teams, and legal departments should never feed draft specifications, confidential invention disclosures, or prosecution strategies into commercial LLMs without enterprise-grade data protection terms. Besides, it is advisable to adjust the filing and prosecution strategies to navigate this evolving examination environment effectively:
- Draft with High Technical Precision: Ensure complete specifications include granular fall-back positions, detailed working examples, and clear functional definitions to pre-empt automated feature-extraction errors.
- Audit Search Logic Early: If an FER cites non-analogous prior art, evaluate whether the underlying classification or search terms were artificially broadened by an automated model. Challenge improper search directions directly in the FER response.
- Rigorously Verify Official Citations: Never assume a polished, well-structured objection in an FER is substantively sound. Cross-check every cited paragraph, translated disclosure, and legal authority against primary sources.
- Deploy Feature-by-Feature Mapping: Counter automated claim splitters with clear, human-analysed claim charts that highlight essential relational limitations that automated tools routinely overlook.
Looking Ahead
The CGPDTM’s guidelines strike a mature balance by embracing technological efficiency while firmly protecting the integrity of legal decision-making. As the Indian Patent Office increasingly integrates automated workflow assistance, success in patent prosecution will depend on vigilance, rigorous source verification, and to-the-point techno-legal arguments.
Authors: Manisha Singh and Joginder Singh