Patent Filing in the AI Era: What Inventors Should Document Before They Apply

Introduction : A concept remains legally inert until it is reduced to a “definite and practical shape” that the Patent Office can accept as proof of conception and reduction to practice. This is a non-negotiable threshold under the Patents Act, 1970, which mandates that the “true and first inventor” must be a natural person, i.e. a human being whose “discernible personal intellectual contribution” is the bedrock of the application. The Indian Patent Office’s Revised Guidelines for Examination of Computer-Related Inventions, 2025, clearly distinguish AI-assisted inventions from AI-generated inventions where the former are patentable and the latter are not because no human inventor can be identified.

This blog discusses the specific records and forms required by the Patents Act and Rules, the record-keeping discipline that patent practice considers good evidence, how the use of AI tools changes what must be documented and a checklist for startups that use AI in product development and invention capture.

Legal Provisions

Who May Apply and Time Frame : Section 6(1) of The Patents Act, 1970 entitles only a person and his legal representative who claims to be the true and first inventor to file a patent application for an invention. Section 7 further states that an application for a patent can be only for a single invention and must be filed in the prescribed form.

Form of the Application and Specification : Rule 13 of The Patents Rules, 2003 states that all the specifications shall be filed in Form 2. Section 10 of the Act, read with Rule 15 and 16, enlists the specifications which are required to be filed along with the application, which consists of a description of the invention along with a suitable title, drawings, a model or sample illustrating the invention if required, the operation or use of the invention and a declaration as to the inventorship. Section 9(1) of The Act provides 12 months, in cases where a provisional specification is provided, to file a complete specification from the date of filing of the application; failing which the application is deemed abandoned.

Section 3(k) and the CRI Guidelines, 2025 : Section 3(k) of the Act excludes mathematical, business methods or computer programmes per se, and algorithms from the scope of patentability. The CRI Guidelines, 2025, published by the Office of the Controller General of Patents, Designs, and Trade Marks, establish a structured technical-effect test for AI and ML claims, stating that a claim survives Section 3(k) only if the specification demonstrates an apparent technical effect, such as faster processing or improved control of a physical process, rather than an abstract algorithmic improvement. The Guidelines further stipulate a clear distinction between AI-assisted inventions and AI-generated inventions. The former are eligible for protection if they meet the requisite patentability criteria and show a discernible human intellectual contribution and the latter are not patentable because AI cannot be recognised as a “person” under Section 6 of the Patents Act.

Legal Analysis

What Must Be Preserved and in What Form : Patent law does not merely ask inventors to remember what happened; it asks them to prove it, and Indian patent practice has developed a specific standard for that proof, commonly called the record book or laboratory notebook, reflecting settled practice built around Sections 2(1)(y), 6 and 7. The recognised practice is to keep entries in a bound, serially numbered book rather than in loose sheets, with the owner's or department’s name and an index of project titles, experiments, results, dates, and page numbers. Each experiment begins on a fresh page and no page is left blank. Even if one is, it is struck through, signed, dated and a reason for the same is recorded. Abandoned experiments carry a noted reason and instructions to assistants and each result, whether success or failure, is recorded with dates, along with the literature considered.

Critically, every page is signed by the person who did the work, countersigned by the project head and witnessed by someone who understands the subject matter well enough to confirm they read and understood the entry and not merely notarised. A notary who has not read the content adds little evidentiary value as the witness must be knowledgeable in the field and not an immediate family member, so the entry is credible if a dispute over inventorship reaches the Controller under Section 28 or a court. Any change in project personnel is also dated and recorded, since this affects who is later named as inventor.

Documents to Prepare and File, and When : Beyond the record book, several formal filings translate that record into the application. Form 1 carries the inventor’s declaration under Paragraph 9; all true and first inventors must be named here at filing, even if signatures follow within six months under Rule 10. Form 2 is the first page of both provisional and complete specifications, distinguished only by its preamble; a provisional specification, filed at conception, secures a priority date without finished claims, while the complete specification, due within twelve months under Section 9(1), must meet the full disclosure standard of Section 10(4). Form 5, the Declaration of Inventorship under Section 10(6), is the only mechanism for adding inventors who contributed after the provisional filing, since Form 1 names cannot be deleted.

How AI-Assisted Work Changes What Needs to Be Recorded : The CRI Guidelines, 2025 do not create a new exemption for AI-assisted inventions but they raise the evidentiary bar for the inventorship and sufficiency requirements already in Sections 6, 7 and 10(4). Two consequences follow. First, since inventorship is reserved for natural persons, the record book must show, for each claimed feature, which human decision, a design choice, a parameter selected, a result validated, produced the inventive step, as distinct from what an AI tool merely proposed. Second, since the technical-effect test requires demonstrable improvement, the underlying data, benchmarks, processing-time comparisons and accuracy figures must be preserved as raw data, since it ultimately supports the best-method and sufficiency requirements of Section 10(4). In practice, the AI tool becomes a specific record-book entry stating what tool was used, what input produced what output, and what the human inventor did with it.

How These Records Support Drafting, Ownership and Prosecution : Strong internal documentation, specifically an authenticated R&D Record Book, provides the foundational evidence required to navigate the entire patent lifecycle for AI-assisted inventions. For drafting, these meticulous logs allow patent practitioners to transcend “theoretical constructs” by capturing specific technical details, such as neural network architecture, data pre-processing logic, and iterative human prompting, that satisfy the “Sufficiency of Disclosure” mandate and anchor the invention to a concrete technical effect. In terms of ownership, records are essential for determining the date of conception and the “true and first” natural person inventors, which is crucial for establishing employer rights in R&D settings and making sure the patent is not deemed ab initio invalid due to improper inventorship. Finally, during prosecution, strong internal records provide the comparative performance data and benchmarking results necessary to rebut Section 3(k) objections, allowing inventors to prove that their AI-assisted method provides a technical solution to a technical problem rather than acting as a mere “algorithm per se”.

Relevant Case Laws

Biswanath Prasad Radhey Shyam vs Hindustan Metal Industries Ltd. (1979) 2 SCC 511 :  The Supreme Court ruled that “obviousness” must be evaluated rigorously and impartially, determining whether an alleged discovery is so far off the known path that it would not naturally occur to a person considering the matter. According to the Court, a simple “workshop result” produced by a skilled craftsman utilising common knowledge without applying creative ingenuity does not qualify as a patentable inventive step.

Ferid Allani vs Union of India & Ors. 2019 SCC OnLine Del 11867 : The Delhi High Court transformed the examination of computer-related inventions by adopting the “technical effect” and “technical contribution” test. The Court clarified that the suffix “per se” in Section 3(k) is not a blanket bar and if a software-based invention demonstrates a discernible technical effect, it surmounts the statutory exclusion and is eligible for patent protection.

Caleb Suresh Motupalli v. Controller of Patents 2025:MHC:293: In addressing the “black box” trap of AI inventions, the Madras High Court ruled that an application fails the enablement test under Section 10(4) if the specification only “coalesces” existing literature without offering specific, useful adoption guidelines. This historic decision requires inventors to go beyond “theoretical constructs” and reveal practical standards that a hypothetical group of experts can replicate without excessive testing.

Practical Implications

For startups using AI tools in product development and invention capture, the risk is not that AI involvement makes an invention unpatentable which the CRI Guidelines, 2025 confirm it does not. Section 6 of the Patents Act along with the 2025 CRI Guidelines expressly prohibit AI from being named an inventor because it is not a legal “person”, so obtaining a patent for an AI-assisted invention is contingent upon adherence to the “Natural Person” mandate. Inventors must move beyond using AI as an unexplained “black box” and instead provide clear, practical directions for its implementation.

Checklist for Startups Using AI Tools in Product Development and Invention Capture

1. Open a dedicated, dated invention record per project from day one: a bound or access-controlled digital equivalent, one experiment or step per entry, mirroring the record-book discipline used to establish conception dates.

2. Log every AI tool interaction feeding a technical result: the tool, the prompt, the raw output, and what the human inventor did with it, rejected, modified or validated.

3. Keep human-directed experimental data separate from AI-suggested output.

4. As required by Section 3(k) of the CRI Guidelines, 2025 and Section 10(4) of the specification, maintain raw technical-effect data rather than summaries, such as processing time, memory or accuracy figures.

5. Have entries witnessed by someone technically competent so it holds weight if inventorship is disputed.

Conclusion

The CRI Guidelines, 2025 and the Patents Act, 1970 read with the Rules of 2003 together settle the doctrinal question: AI-assisted inventions remain patentable, AI-generated ones do not and the difference turns on whether a human inventor’s contribution can be shown. The real work of “documenting an invention” happens months before anyone drafts a claim. It happens the day someone opens a record book, decides an AI tool’s suggestion is worth pursuing and writes down what they changed about it. Every stage that follows filing, examination, opposition and enforcement depends on evidence that has to already exist by the time it is needed. Every design decision, every test result, every moment where a human inventor accepted, modified or rejected an AI tool’s suggestion is potential evidence of the intellectual contribution that patent law requires. Thus, it is significant to file the record before one files the application, as one protects the other.

Author :- Damita, in case of any query, contact us at Global Patent Filing or write back us via email at support@globalpatentfiling.com.

Endnotes

1. The Patents Act, 1970, No. 39, Acts of Parliament, 1970 (India).
2. The Patents Rules, 2003 (India).
3. The Patents (Amendment) Rules, 2024, G.S.R. 211(E) (India).
4. Office of the Controller Gen. of Patents, Designs and Trade Marks, Guidelines for Examination of Computer Related Inventions (CRIs) (2025) (India).
5. Office of the Controller Gen. of Patents, Designs and Trade Marks, Guidelines for Search and Examination of Patent Applications (2015) (India).
6. Biswanath Prasad Radhey Shyam v. Hindustan Metal Indus., AIR 1982 SC 1444 (India).
7. Ferid Allani v. Union of India, 2019 SCC OnLine Del 11867 (India).
8. Caleb Suresh Motupalli v. Controller of Patents, 2025:MHC:293 (India).
9. Feroz Ali Khader, The Law of Patents (LexisNexis 2026).
10. N.R. Subbaram, Patent Law Practice & Procedures (LexisNexis 2026).