On August 7, 2026, the Office of The Controller General Of Patents, Designs And Trade Marks (O/o CGPDTM) published a sixteen-page policy document titled “Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures,” along with two detailed annexures running to a further eighteen pages. The Guidelines set out, for the first time, a structured framework governing how patent examiners and controllers may use AI tools, including generative AI, while carrying out their statutory functions under the Patents Act, 1970.
This is a different question from the one Indian patent law has already spent the last year working through. The CGPDTM’s Guidelines for Examination of Computer Related Inventions, issued in July 2025, addressed whether AI-related inventions can be patented. These new Guidelines address something else entirely: whether, and how, the examiners themselves can use AI as a tool while doing their jobs.
What the Guidelines actually cover
The scope is defined broadly. The Guidelines apply to the use of AI in “screening, classification, search, translation support, drafting support, technical comparison or knowledge retrieval,” and they distinguish between public AI tools (consumer-facing platforms trained on broad public datasets) and private AI tools (proprietary systems, including the AI search tools already built into the databases the Patent Office subscribes to for prior art search).
The document’s central position is stated plainly: AI is meant to assist the examiner, not replace their judgment. Any AI-assisted output used in the course of examination must be independently reviewed and verified before it is relied upon, and the examiner or controller remains fully responsible for the final outcome regardless of what tools were used to get there.
To make this concrete, the Guidelines set out twelve specific categories of permitted use, each paired with its own risk profile and required safeguards. These range from relatively low-stakes tasks, using AI to suggest patent classification codes or generate prior-art search vocabulary, to more sensitive ones, using AI for preliminary novelty or inventive-step analysis, or to identify possible clarity issues in a claim. In every category, the safeguard is the same in substance: the officer must independently verify the output against primary sources before it can factor into an actual decision.
Five uses are listed as prohibited outright. These include entering unpublished application content or confidential office material into public AI tools, using AI as a substitute for the examiner’s own application of mind on questions like novelty or inventive step, issuing office actions or hearing notices based solely on AI-generated output, and relying on AI-suggested case law or prior art without independently verifying it against the original source.
The worked examples
What sets this document apart from a typical policy circular is that it doesn’t just state the risks in the abstract, it demonstrates them with real, worked examples, several of which are genuinely useful reading for anyone trying to understand how generative AI tools behave in a technical, high-stakes setting.
One example tracks what happened when an AI tool was asked to provide “verbatim quotes” on the legal concept of “teaching away,” drawing on a 2020 Intellectual Property Appellate Board decision, Pharmacyclics, an AbbVie Company v. Controller General of Patents, concerning a patent for the drug Ibrutinib. The AI tool produced several precisely formatted quotations, complete with paragraph-level citations, presented as direct extracts from the judgment. When asked to provide the source, it offered a citation. Only when directly challenged a third time did the tool acknowledge that the quotations were not verbatim at all, but its own reconstructions based on general knowledge of the case. The Guidelines use this sequence as a worked illustration of what they term “fabricated verbatim quotations,” a specific and, in their words, particularly serious form of AI hallucination, and the reasoning behind the rule that no AI-suggested citation may be used in examination without being checked against the actual primary source.
A second example compiles results from several different AI tools asked to suggest patent classification codes for the same two patent applications. The outputs varied substantially between tools, both in the number of codes suggested and in how closely they matched the classification ultimately assigned through manual review, with some tools returning codes the Guidelines note were not relevant to the claimed invention at all. The exercise is presented as evidence that classification and search outputs are, in the document’s own terminology, “model-sensitive,” and that this variability is itself the reason independent verification remains necessary regardless of which tool is used.
Administrative structure going forward
The Guidelines also set up the governance structure meant to keep this framework current. A dedicated AI Governance Committee is to be established within the CGPDTM, drawing examiners and controllers from the Examination Division alongside officers from the IT Office and Quality Management System Division, with responsibility for approving tools, reviewing errors or policy breaches, and periodically revising the Guidelines as the underlying technology changes. The document also anticipates training programs for officers covering AI’s functional limitations, prompt discipline, and how to recognise hallucinated or fabricated output, along with the possibility of independent audits and incident reporting mechanisms.
The Guidelines describe themselves as informed by the Indian government’s broader responsible AI policy work, including NITI Aayog‘s Responsible AI approach documents, and situate the patent examination context, described as “document-intensive, technically complex, legally evolving and time-sensitive,” as one where the benefits of AI assistance are real but where the underlying decisions remain quasi-judicial in nature.
Why this is worth reading closely, not just knowing about
For anyone currently preparing for the Patent Agent Examination, or working toward a career inside the patent system, this document is a useful preview of the profession as it currently stands, not a hypothetical about AI’s future role, but a live, dated policy already shaping how examiners are expected to work today. The permitted-use categories in particular map closely onto the same tasks, classification, prior art search, claim analysis, drafting, that any patent agent or examiner performs day to day, which makes the Guidelines as much a snapshot of the job itself as a statement of AI policy.
Source: Office of the Controller General of Patents, Designs and Trade Marks (CGPDTM), “Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures,” published August 7, 2026, available via the official notice on ipindia.gov.in. All examples and figures in this piece are drawn directly from the published Guidelines and its annexures; nothing has been reproduced verbatim beyond the short illustrative excerpts quoted above.
