BEYOND THE HUMAN MIND: A CRITICAL ANALYSIS OF INVENTORSHIP AND SUFFICIENCY OF DISCLOSURE FOR AI-GENERATED INVENTIONS UNDER THE INDIAN PATENTS ACT, 1970

I. Introduction

For most of the patent system’s history, the machine has been a tool and the human has been the inventor.[1] A scientist using a mass spectrometer to identify a compound is still the inventor of that compound; the instrument merely extends the reach of human perception.[2] Generative artificial intelligence disturbs this settled division of labour.[3] Models trained on protein structures, reaction pathways, or structural load data can now propose novel molecules and engineering configurations that no human researcher specifically directed, anticipated, or even fully understands at the moment of generation.[4] The shift is not merely one of degree but of kind: where earlier software assisted a human inventor in executing an idea that had already been conceived, contemporary generative systems increasingly perform the act of conception itself, leaving the human role confined to framing the problem, curating training data, or selecting among outputs.[5]

This shift creates two distinct but related legal difficulties for a patent regime built on the assumption of a human “first and true inventor.”[6] The first is a question of entitlement: when the inventive contribution traceable to a human is minimal or absent, who may lawfully claim to be the inventor, and through what chain of title does a corporate or individual applicant derive the right to apply for a patent at all?[7] The second is a question of disclosure: patent law trades a temporary monopoly for public knowledge, and that trade depends on the applicant being able to explain, in terms intelligible to a notional skilled person, how the invention works and how it may be reproduced.[8] Where the generative model operates as a “black box,” producing outputs through processes that are not fully traceable even by its own designers, the applicant may be unable to satisfy this disclosure obligation in the manner the statute contemplates.[9]

The Indian Patent Office’s refusal of Stephen Thaler’s DABUS applications on 15 April 2026 confirms that India has aligned itself with the global consensus that an artificial intelligence system cannot be named as an inventor.[10] That alignment, however, resolves only the threshold question of nomenclature.[11] It leaves open the harder and more consequential questions of entitlement and disclosure that arise once a human applicant, rather than the machine, stands at the centre of the application.[12] This article examines those questions under the existing text of the Patents Act, 1970, evaluates the comparative jurisprudence that has shaped India’s position, and proposes a calibrated set of administrative and statutory responses that neither anthropomorphise the machine nor ignore the practical reality of how such inventions are now made.[13]

II. The Inventorship Crisis: Reading Sections 2(1)(j), 6 and 10(6) Together

The starting point is definitional.[14] Section 2(1)(j) of the Patents Act, 1970 defines an “invention” as a new product or process involving an inventive step and capable of industrial application.[15] Section 2(1)(ja) in turn defines “inventive step” as a feature that involves technical advance, or economic significance, or both, and that renders the invention not obvious to a person skilled in the art.[16] Neither provision speaks directly to the identity of the inventor, but both presuppose an evaluative exercise conducted from the vantage point of a hypothetical skilled person assessing obviousness as of a given date.[17] That framework was drafted with a human conceiver in mind: obviousness is measured against what a person of ordinary skill would have arrived at, which is a comparison between two minds, not between a mind and a statistical model.[18]

The Act addresses inventorship more directly, if obliquely, through its definitional and procedural provisions.[19] Section 2(1)(y) defines “true and first inventor” only in the negative, excluding from that category the first importer of an invention into India and a person to whom an invention is first communicated from abroad.[20] The term itself is left otherwise undefined, on the apparent assumption that its ordinary meaning, a person who conceives an invention, requires no further elaboration.[21] Section 6 then restricts the class of persons who may apply for a patent to the true and first inventor, the assignee of that inventor’s right to apply, or the legal representative of a deceased inventor.[22] Each of these three categories presupposes a legal person capable of holding rights, of executing an assignment, or of leaving an estate to be represented.[23] An artificial intelligence system, however sophisticated its output, does none of these things.[24]

The procedural provisions reinforce this structural reading.[25] Section 7(2) permits a person other than the true and first inventor to be named as inventor only where that person or the applicant makes the relevant declaration in the prescribed manner.[26] Section 7(3) requires the application to name the person claimed to be the true and first inventor and, where the applicant is not that person, to declare a belief that the named person is indeed the true and first inventor.[27] Section 10(6) requires a declaration as to inventorship to accompany the complete specification in the prescribed form.[28] None of these provisions can sensibly be executed by a non-human system: a declaration is a legal act requiring an entity capable of holding a belief and bearing responsibility for its truth, and an assignment under Section 6(b) requires an assignor capable of holding the underlying right in the first instance.[29]

This is precisely the structural objection that the Indian Patent Office adopted in its twin orders of 15 April 2026 on the DABUS application.[30] The Controller held that the scheme of Sections 6, 7 and 10 necessarily presupposes an inventor capable of holding legal rights, executing a declaration, and transferring those rights to an assignee, and that a machine can do none of these things.[31] The Office went further and rejected Thaler’s alternative theory that ownership of the AI system, its source code, and the computer on which it operated could itself confer entitlement to apply.[32] Ownership of the instrument, the Controller reasoned, does not create a chain of title under Section 7(2) unless that chain originates from a legally recognisable inventor; owning a machine is not the same as deriving a right from a person.[33]

The doctrinal force of this reasoning is considerable, but it does not by itself answer the ownership question that arises once the human role in an AI-generated invention falls below the threshold that would ordinarily support a claim to inventorship.[34] If the system, rather than any identifiable researcher, supplies the inventive concept, and if Indian law requires a human “true and first inventor” as a precondition to any valid application, two outcomes seem available: either the application must identify a human contributor whose role, however modest, can be characterised as the conception of the invention, with the consequence that questionable inventorship claims may be advanced simply to satisfy a formal requirement, or the invention is rendered effectively unpatentable in India regardless of its novelty, utility, or industrial value.[35] Neither outcome is satisfactory.[36] The first risks converting inventorship from a substantive inquiry into a fiction maintained for administrative convenience, undermining the integrity of the declaration that Section 7(3) requires.[37] The second risks leaving a growing category of genuinely valuable technical advances outside the patent system entirely, with consequences for investment in AI-driven research and development that the legislature has not yet had occasion to weigh.[38]

III. The Black-Box Dilemma: Section 10(4) and the Limits of Explicability

Even where a human applicant can properly be named as the inventor, whether as the person who trained, directed, or curated the AI system’s inputs, or as the person who reduced its output to a workable specification, a second and independent obstacle arises at the disclosure stage.[39] Section 10(4) sets out the contents required of every complete specification.[40] It must fully and particularly describe the invention, its operation or use, and the method by which it is to be performed; it must disclose the best method of performing the invention known to the applicant; it must end with a claim or claims defining the scope of protection sought; and it must be accompanied by an abstract conveying technical information about the invention.[41]

The Intellectual Property Appellate Board has read this requirement generously in conventional cases, holding that sufficiency is satisfied where at least one way of working the invention is clearly described so as to enable a skilled person to carry it out, and that an applicant need not enable every conceivable variant.[42] The Delhi High Court has nonetheless insisted, most recently in its review of an application by the Regents of the University of California, that the description must particularly explain the manner in which the invention is to be performed and that claims unsupported by an adequate description will not survive scrutiny.[43] The animating principle, recognised across this body of case law, is the quid pro quo at the heart of the patent bargain: the applicant receives a time-limited monopoly in exchange for placing sufficient knowledge in the public domain to permit replication once that monopoly expires.[44]

Generative AI systems strain this bargain in a way that earlier computer-implemented inventions did not.[45] A conventional algorithm, however complex, can in principle be described step by step, because its designer wrote each step.[46] A deep generative model that proposes a novel molecular structure or an engineering configuration arrives at its output through the interaction of millions or billions of learned parameters operating over a multidimensional space, a process that is not merely lengthy to describe but is, in important respects, not fully known even to the system’s own developers.[47] This is the sense in which such systems are properly called “black boxes”: the relationship between the training data, the model architecture, and any specific output is not one that admits of a step-by-step causal account in the way that Section 10(4)(a) appears to demand.[48]

The applicant facing this difficulty has, broadly, three options, none of which sits comfortably within the existing statutory text.[49] The applicant may describe the final output, the molecule or the engineering system itself, in full structural or functional detail without attempting to explain the generative process that produced it.[50] This approach can satisfy Section 10(4)(a) and (b) as applied to the invention itself, since the provision is concerned with enabling the skilled person to perform the claimed invention, not with requiring a confession of how the applicant arrived at the idea.[51] Difficulties remain, however, where the inventive step itself lies partly in the generative process or where downstream claims depend on reproducing variants that only the model, and not the applicant, can generate.[52] Alternatively, the applicant may attempt to describe the training methodology, architecture, and parameters of the model in general terms, an approach that risks being simultaneously too vague to satisfy a sceptical examiner and too revealing of commercially sensitive trade secrets to be commercially palatable.[53] A third, more technically demanding option involves treating the trained model itself, or a representative dataset and parameter set sufficient to reproduce its outputs, as analogous to the biological material that the proviso to Section 10(4) already permits to be deposited with an authorised depository rather than fully described in words, an analogy considered further in Part V.[54]

IV. Comparative Jurisprudence: The DABUS Line and India’s April 2026 Position

Stephen Thaler’s attempt to register DABUS as the sole inventor of a food container and a search-and-rescue beacon has now been tested across nearly every major patent jurisdiction, producing a remarkably uniform answer to the inventorship question even as the underlying statutory language varies.[55] In the United States, the Court of Appeals for the Federal Circuit held in Thaler v. Vidal that the Patent Act defines an inventor as an individual, a term that Congress has used to denote a natural person, and affirmed the USPTO’s refusal on that basis.[56] The European Patent Office’s Legal Board of Appeal reached the same destination by a different route in its December 2021 decisions on Thaler’s applications, holding that the European Patent Convention requires a designated inventor to be a person with legal capacity, while expressly leaving open the separate question of whether an AI-generated invention could otherwise meet the Convention’s patentability criteria.[57] The United Kingdom Intellectual Property Office’s refusal, upheld on further appeal, rested on the parallel ground that ownership of an AI system could not substitute for inventorship as a matter of factual contribution, since inventorship under the relevant provisions turns on who in fact devised the invention rather than on who controls the apparatus that produced it.[58] Australia presents a partial complication: a single judge of the Federal Court initially accepted Thaler’s argument that nothing in the Patents Act 1990 (Cth) expressly excluded a non-human inventor, but the Full Court of the Federal Court set that reasoning aside in 2022, holding that only a natural person can be an inventor for the purposes of the Australian Act, a position the High Court declined to revisit on special leave.[59]

India’s own resolution of the question, delivered through the Patent Office’s twin orders of 15 April 2026 on application 202017019068, sits comfortably within this comparative consensus rather than departing from it.[60] The Controller’s reasoning drew directly on the structural requirements of Sections 6, 7 and 10 discussed in Part II, and the order expressly noted the persuasive force of the Australian Full Court’s interpretation of an analogous entitlement provision.[61] The Indian decision, like its foreign counterparts, did not rest inventorship on philosophical claims about machine creativity or consciousness; it rested on the narrower, more tractable ground that the statutory scheme requires an inventor capable of bearing legal rights and obligations, a threshold a software system cannot cross regardless of the sophistication of its output.[62]

What distinguishes the Indian order, and what offers a more promising avenue for future AI-assisted applicants than the DABUS litigation itself, is its evident reluctance to foreclose patentability for inventions in which an AI system plays a substantial but not exclusive role.[63] The Assistant Controller’s observation that, because inventive step is assessed from the standpoint of a person skilled in the art, the conception of an invention should also originate from a human mind, is framed as a statement about DABUS’s particular claim to sole inventorship rather than as a categorical rule against any AI involvement in the inventive process.[64] This leaves room, consistent with practice that has already developed around AI-assisted drug discovery and engineering design, for applications in which a human researcher who directs the system, selects among its outputs, or reduces a generated candidate to a workable invention is named as the true and first inventor, even where a generative model performed a significant part of the underlying exploratory work.[65] The doctrinal question that India has not yet had to confront directly, and that is likely to dominate the next phase of this debate, is how much human involvement is enough, a question on which the inventive-step jurisprudence under Section 2(1)(ja) offers only an indirect and as yet untested guide.[66]

V. Policy Recommendations: Toward a Calibrated Disclosure-Ownership Framework

The comparative consensus against AI inventorship is, on balance, the correct reading of the existing statutory text, and there is little to be gained from urging the Patent Office or the courts to depart from it through interpretation alone.[67] The more productive task is to address the two practical difficulties that the DABUS line leaves unresolved: providing a workable and honest basis for ownership claims where human inventive contribution is minimal, and adapting the disclosure regime to the realities of generative model outputs without diluting the quid pro quo that justifies patent protection in the first place.[68] Four proposals follow:[69]

1. AI Contribution Statement: The Patent Rules, 2003 might usefully be amended to require an “AI Contribution Statement” accompanying Form 1 and the declaration under Section 7(3), in which the applicant discloses, without altering the legal requirement of a named human inventor, the extent to which a generative or other AI system contributed to the conception of the claimed invention.[70] This would not create a new category of inventorship; it would simply make visible, for the Controller’s and the public’s benefit, information that currently exists in no formal channel and that bears directly on whether the named human inventor’s declaration under Section 7(3) reflects an honest account of how the invention was conceived.[71] Such a requirement would also generate, over time, the body of administrative practice that Indian courts presently lack when assessing borderline claims of human inventorship in AI-assisted contexts.[72]

2. Defined Human Involvement Threshold: The threshold question of how much human involvement suffices to support inventorship should be addressed expressly, whether by Patent Office guidelines or, more durably, by an amendment to Section 2, rather than left to accumulate through inconsistent first-instance practice.[73] A workable formulation, drawing on the structure already implicit in the Assistant Controller’s April 2026 order, would treat a human as the true and first inventor where that person formulated the technical problem to be solved, exercised a non-trivial selective or evaluative judgment over the system’s candidate outputs, and possessed the technical understanding necessary to recognise why a given output solved the problem.[74] This formulation excludes the case, illustrated by DABUS itself, in which the human’s role is limited to owning and maintaining the system, while accommodating the much more common scenario in which a researcher directs an AI tool toward a defined objective and exercises genuine judgment in selecting and validating its output.[75]

3. Model Deposit Mechanism: Section 10(4) could be adapted, for the narrow category of inventions substantially shaped by generative AI systems, along lines analogous to the existing proviso permitting deposit of biological material with an international depository authority under the Budapest Treaty where written description alone cannot adequately capture the material.[76] Just as that proviso recognises that some subject matter resists full verbal description and substitutes a deposit-and-access mechanism while preserving the disclosure function, a parallel mechanism could permit an applicant to satisfy sufficiency by depositing, with a recognised technical authority, a defined snapshot of the model, training data summary, or parameter set sufficient to allow a suitably resourced skilled person to reproduce the relevant output, while the specification itself continues to fully and particularly describe the resulting product or process as Section 10(4)(a) already requires.[77] This avoids the false choice between an unworkably literal demand for step-by-step algorithmic explanation and a hollowed-out disclosure obligation that describes only the end result.[78]

4. Heightened Evidentiary Standard for Inventive Step: Examiners assessing inventive step in this category of application would benefit from a heightened evidentiary standard modelled on the approach the Supreme Court took toward efficacy claims in Novartis AG v. Union of India.[79] The Court’s insistence in that case on rigorous, evidence-based substantiation of claimed therapeutic efficacy, rather than acceptance of the applicant’s characterisation at face value, offers a useful template for AI-generated inventions, where the risk of overstated or under-substantiated claims to technical advance is at least as great.[80] An applicant relying on AI-assisted generation to support a claim of non-obviousness under Section 2(1)(ja) should be expected to substantiate, with comparative data against the closest prior art, why the specific output represents a genuine technical advance rather than a predictable variation that the model’s scale and search capacity happened to surface.[81]

None of these proposals requires recognising an AI system as a legal person or as an inventor.[82] Each instead accepts the structural premise that Sections 6, 7 and 10 already establish, that inventorship and the right to apply are matters for legal persons alone, while building the administrative and statutory infrastructure needed to apply that premise honestly and workably to a generation of inventions in which the human contribution, though real and legally necessary, is no longer the whole story.[83]

VI. Conclusion

The Indian Patent Office’s refusal of the DABUS application confirms what the comparative jurisprudence already made evident: under the text of the Patents Act, 1970, an artificial intelligence system cannot be the true and first inventor, because the statutory scheme of inventorship, declaration, and assignment presupposes a legal person capable of holding rights and bearing responsibility for the truth of a declaration.[84] That conclusion, however, settles only the easiest of the questions that generative AI poses to Indian patent law.[85] The harder questions, of how much human involvement is enough to ground an honest claim to inventorship, and of how a disclosure regime built for step-by-step human reasoning should treat inventions substantially shaped by processes that resist step-by-step explanation, remain open.[86] The proposals advanced here—an AI contribution statement, a clearer threshold for human inventive contribution, a deposit-based disclosure mechanism modelled on the existing biological material proviso, and a Novartis-style evidentiary standard for inventive step—would allow Indian law to maintain its settled and, in this author’s view, correct position on inventorship while still accommodating the genuine technical advances that AI-assisted research is now capable of producing.[87] The alternative, treating the DABUS refusal as the end of the inquiry rather than its beginning, risks leaving Indian patent law increasingly disconnected from the way invention itself is coming to be practised.[88]

References

[1] Priya Ray, Beyond the Human Mind: A Critical Analysis of Inventorship and Sufficiency of Disclosure for AI-Generated Inventions under the Indian Patents Act, 1970, Nat’l Law Univ. Delhi (2026), Section I.
[2] Id.
[3] Id.
[4] Id.
[5] Id.
[6] The Patents Act, No. 39 of 1970, INDIA CODE (1970), § 2(1)(y).
[7] Ray, supra note 1, Section I.
[8] The Patents Act, 1970, § 10(4).
[9] Ray, supra note 1, Section I.
[10] Indian Patent Office Order on Application No. 202017019068 (DABUS) (15 April 2026).
[11] Ray, supra note 1, Section I.
[12] Id.
[13] Id.
[14] Ray, supra note 1, Section II.
[15] The Patents Act, 1970, § 2(1)(j).
[16] Id. § 2(1)(ja).
[17] Ray, supra note 1, Section II.
[18] Id.
[19] Id.
[20] The Patents Act, 1970, § 2(1)(y).
[21] Ray, supra note 1, Section II.
[22] The Patents Act, 1970, § 6.
[23] Ray, supra note 1, Section II.
[24] Id.
[25] Id.
[26] The Patents Act, 1970, § 7(2).
[27] Id. § 7(3).
[28] Id. § 10(6).
[29] Ray, supra note 1, Section II.
[30] Indian Patent Office Order on Application No. 202017019068 (15 April 2026).
[31] Id.
[32] Id.
[33] Id.
[34] Ray, supra note 1, Section II.
[35] Id.
[36] Id.
[37] Id.
[38] Id.
[39] Ray, supra note 1, Section III.
[40] The Patents Act, 1970, § 10(4).
[41] Id. § 10(4)(a)–(d).
[42] Ray, supra note 1, Section III.
[43] Regents of the Univ. of Cal. v. Controller Gen. of Patents, Designs & Trademarks, (Delhi High Court).
[44] Ray, supra note 1, Section III.
[45] Id.
[46] Id.
[47] Id.
[48] Id.
[49] Id.
[50] Id.
[51] The Patents Act, 1970, § 10(4)(a)–(b).
[52] Ray, supra note 1, Section III.
[53] Id.
[54] Id.; Budapest Treaty on the International Recognition of the Deposit of Microorganisms for the Purposes of Patent Procedure, Apr. 28, 1977, 1861 U.N.T.S. 361.
[55] Ray, supra note 1, Section IV.
[56] Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022).
[57] Legal Board of Appeal of the European Patent Office, Decisions J 0008/20 & J 0009/20 (21 Dec. 2021).
[58] Thaler v. Comptroller-General of Patents, Designs and Trade Marks, [2023] UKSC 49.
[59] Commissioner of Patents v. Thaler, [2022] FCAFC 62 (Australia).
[60] Indian Patent Office Order on Application No. 202017019068 (15 April 2026).
[61] Id.
[62] Ray, supra note 1, Section IV.
[63] Id.
[64] Indian Patent Office Order, Application No. 202017019068 (15 April 2026).
[65] Ray, supra note 1, Section IV.
[66] Id.
[67] Ray, supra note 1, Section V.
[68] Id.
[69] Id.
[70] The Patent Rules, 2003 (India), Form 1 & Rule 13; Patents Act, 1970, § 7(3).
[71] Ray, supra note 1, Section V.
[72] Id.
[73] Id.
[74] Id.
[75] Id.
[76] The Patents Act, 1970, § 10(4) proviso.
[77] Id. § 10(4)(a).
[78] Ray, supra note 1, Section V.
[79] Novartis AG v. Union of India, (2013) 6 SCC 1.
[80] Id.; Ray, supra note 1, Section V.
[81] The Patents Act, 1970, § 2(1)(ja).
[82] Ray, supra note 1, Section V.
[83] Id.
[84] Id. Section VI; Indian Patent Office Order on Application No. 202017019068 (15 April 2026).
[85] Ray, supra note 1, Section VI.
[86] Id.
[87] Id.
[88] Id.