Augmented inventorship: rethinking patent law in the age of generative AI

Abstract

Generative artificial intelligence (AI) now participates in tasks constitutive of invention—problem framing, hypothesis generation, and design—yet patent doctrine remains anchored to a natural-person rule that offers limited guidance for AI-intensive workflows. This Article advances augmented inventorship, a conservative but operationally modern attribution doctrine that preserves human inventorship while making AI’s generative role legible and auditable at the moment of conception. Drawing on an analogy to augmented immunology, the framework identifies two design criteria—directability (independent and substantive human intellectual judgment steering model behavior or selection) and traceability (a reviewable, claim-centered record linking human reasons to claim elements)—and translates them into a proportionate evidentiary practice: a Computational Traceability Report and a Human–Machine Contribution Statement. These instruments are content-rich but code-light. They support enablement and sufficiency, clarify claim drafting and construction, reduce prosecution and litigation error costs, and balance evidentiary transparency with trade-secret sensitivity through proportional disclosure. Situated within—and distinguished from—the growing literature on AI inventorship and disclosure, the doctrine aligns with existing law (US conception and significant-contribution standards; the UK’s ‘actual deviser’; EPC sufficiency) and is compatible with TRIPS disclosure norms. Rather than demanding ‘more disclosure’ in the abstract, augmented inventorship supplies an administrable grammar for human accountability in AI-assisted research.

I. INTRODUCTION

Generative artificial intelligence (AI) has altered both the epistemology and the practice of invention.¹ State-of-the-art deep-learning systems, including transformer architectures, now perform tasks constitutive of inventive activity—framing problems, generating and ranking hypotheses, designing molecules and materials, and executing automated design–build–test–learn cycles across combinatorial spaces beyond unaided human capacity.² In the biosciences, structure-prediction and design platforms have reshaped discovery by producing candidate solutions that plausibly satisfy novelty, non-obviousness/inventive step, and industrial applicability.³ Outputs increasingly emerge from human–machine workflows in which model-driven generativity and human judgment are interdependent.

Patent law remains organized around a human-centered theory of conception and a natural-person rule for inventorship. Across major forums, current law does not accommodate a non-human inventor: the UK Supreme Court, the European Patent Office’s (EPO’s) Legal Board of Appeal, and the US Court of Appeals for the Federal Circuit have each so held.⁴ In parallel, the USPTO’s 2024 guidance frames inventorship in AI-assisted cases around a ‘significant human contribution’ to conception.⁵ In the UK, inventorship turns on the ‘actual deviser’ under the Patents Act 1977, s 7(3), as read in Yeda; under the EPC, the parallel disclosure duty of sufficiency (Art. 83) ensures the skilled person can practice the claimed invention across its scope.⁶ These determinations preserve doctrinal cores—humans swear oaths, bear duties, and assign rights—but they do not, by themselves, specify how to govern AI-intensive invention at the evidentiary interface of prosecution and litigation.

The difficulty is concrete rather than abstract. Consider a routine laboratory scenario. A team deploys a generative model trained on curated sequence–function data to propose hundreds of protein variants; top-ranked designs exhibit properties later embodied in the claims. The principal investigator defines the problem, curates the data, sets thresholds, rejects most outputs, and validates a handful experimentally, refining them into a final construct. Who is the inventor, and what must be disclosed? Treating the model as a mere ‘tool’ obscures the human–machine division of labor at the moment of conception, leaving examiners and courts without a principled basis to determine whether there was a significant human contribution,⁷ whether enablement/sufficiency is satisfied when core features depend on opaque transformations, and whether duties of candor or good faith required disclosure of prompts, datasets, or thresholding choices. The resulting opacity increases error costs, distorts obviousness and claim-scope analysis, encourages strategic secrecy, and risks forum shopping.

This article advances a focused refinement: Augmented Invention. The thesis is that patent law should preserve the natural-person baseline while making AI’s generative role legible and auditable. Two commitments structure the proposal. First, epistemic responsibility: inventorship is limited to actors who can understand, justify, and bear legal duties for the invention, even when AI materially contributes. Secondly, traceability: applicants should provide concise, claim-centered documentation mapping human decisions to AI-mediated transformations and to the features later claimed. On this view, AI is neither ignored nor personified; it is a generative collaborator whose role must be evidenced, while inventorship and liability remain human. The normative grounding draws on augmented immunology: as with mRNA vaccination and CAR-T therapy—which amplify endogenous agency without displacing it—AI can augment human inventive capacity without relocating legal accountability.⁸ The contribution is both doctrinal and operational: a named doctrine of Augmented Inventorship and a proportionate evidentiary grammar Computational Traceability Report/Human–Machine Contribution Statement (CTR/HMCS) that maps human reasons to claim elements, making the ‘significant human contribution’ standard administrable across forums. I distinguish AI-assisted research (tool use) from augmented inventorship (the attribution rule for conception in human–machine generativity). Under this framework, inventorship remains human only where articulated reasons steered model behavior or selection so that at least one claim element reflects that judgment, as evidenced by CTR/HMCS.

II. POSITIONING AUGMENTED INVENTORSHIP IN THE SCHOLARLY LITERATURE

Existing debates on AI and inventorship tend to cluster around a limited set of approaches. One dominant view treats AI systems as increasingly sophisticated tools, preserving traditional human inventorship while leaving the evidentiary analysis of conception largely unchanged.⁹ A second approach urges recognition of AI systems themselves as inventors, motivated by descriptive candor where machine-generated outputs appear to satisfy claim-defining features.¹⁰ A third family of proposals seeks to preserve the human-inventor baseline through enhanced disclosure or ownership-based attribution, emphasizing transparency about AI involvement without reconceptualizing inventorship itself.¹¹

Each of these approaches captures an important aspect of AI-assisted research, yet none resolves the central problem that generative systems introduce for patent law in the biosciences: the evidentiary gap at the moment of conception. Treating AI as a mere tool obscures how human judgment interacts with model-driven generativity when claim-defining features are fixed. Recognizing AI as an inventor conflicts with settled statutory and doctrinal commitments to natural-person inventorship and raises intractable questions of legal responsibility, declaration, and assignment. Disclosure-centered proposals rightly emphasize transparency but typically operate at the project or system level, leaving under-specified how examiners and courts should assess whether human intellectual contribution crystallized conception with respect to particular claim elements.

This Article advances a distinct account. Augmented inventorship preserves the natural-person rule while reconceiving inventorship as a claim-centered attribution inquiry under conditions of human–machine generativity. The contribution is not ‘more disclosure’ in the abstract, but a structured evidentiary grammar that ties human reasons to claim elements through traceability at conception. By requiring applicants to show how identifiable human agents framed, steered, evaluated, or selected AI-generated outputs such that at least one claimed feature reflects independent intellectual judgment, the doctrine makes the ‘significant human contribution’ standard administrable across jurisdictions. In AI-intensive bioscience research, where models routinely propose candidates across vast combinatorial spaces, this approach supplies what existing frameworks lack: a principled, auditable method for attributing inventorship without conferring legal status on machines or collapsing inventorship into ownership or control.

III. AUGMENTED IMMUNOLOGY AS A NORMATIVE ANALOGY

The analogy to contemporary immunology clarifies why augmentation—rather than substitution—is the right frame. mRNA vaccination delivers transient instructions that prompt host cells to express antigen, priming adaptive immunity with speed and specificity while clinicians determine indication, dose, and monitoring. CAR-T therapy reprograms a patient’s T cells ex vivo to recognize malignancies; responsibility remains human and clinical, from preparation through post-infusion monitoring, including management of cytokine-release syndrome. In both settings, synthetic inputs expand capability without displacing the locus of responsibility.¹²

Two design criteria for legal attribution follow. Directability captures that augmentation is bounded by ex ante choices—sequence design, delivery vehicle, receptor architecture—and by ex post governance—titration, monitoring, intervention. Traceability is the evidentiary counterpart: biomedical augmentation leaves a structured record—protocols, batch documentation, release specifications, and quality-control metrics—linking design choices to outcomes and making responsibility auditable.

This analogy surfaces a preliminary point about the concept of invention. US doctrine treats invention as conception—a definite and permanent idea enabling practice without undue experimentation—while UK practice asks who was the ‘actual deviser’; EPC law imposes the parallel duty of sufficiency so the skilled person can work the invention across the scope.¹³ Read together, these frames allow AI-intensive research to be governed without abandoning the human-inventor baseline, provided that claim-specific crystallization remains intelligible and attributable to human reasons. Transposed to invention, the task is to maintain the link between control and accountability: did human reasons steer generative processes at claim-defining junctures, and is there a reviewable trail connecting those reasons to the features later asserted? The analogy also illuminates risk management: where immunology manages physiologic hazards through protocol and oversight, invention must manage epistemic risks—spurious correlations, training-data artifacts, overfitting—through structured records evidencing human filtering and stabilization at the point of conception.

IV. THE DOCTRINE OF AUGMENTED INVENTORSHIP

Augmented inventorship preserves the human-inventor baseline while furnishing an attribution rule suitable for AI-intensive research. Current law recognizes only natural persons as inventors; leading forums have read their instruments to exclude non-human inventors.¹⁴ Within that constraint, invention should be understood as a human-anchored, machine-augmented enterprise: model-driven generativity may supply intermediate artifacts or refine designs, but inventorship vests only where identifiable human agents exercise meaningful oversight at the point of conception and can explain, justify, and assume legal responsibility for the features later claimed. In US terms, conception remains the ‘formation … of a definite and permanent idea’ of the complete and operative invention, with significant contribution as the touchstone; in UK terms, inventorship turns on the ‘actual deviser;’ for the EPC, a clear account of how the claimed solution was fixed supports sufficiency across the scope, as underscored in Regeneron v. Kymab.¹⁵

For administrability, meaningful oversight must be tied to claim elements. The threshold is met only where the putative inventor can articulate the reasons that steered model behavior or selection so that at least one claim element reflects that judgment. By contrast, merely running a pre-trained system, performing clerical steps in an automated pipeline, or ratifying whatever the model proposed after the fact is not enough. This keeps inventorship distinct from ownership or managerial control and coheres with both the US significant-contribution requirement and the UK actual-deviser inquiry.¹⁶

The standard is concrete. A principal investigator who (i) defines the binding-site objective, (ii) excludes cross-reactive epitopes during curation, (iii) sets rejection thresholds for generative designs, and (iv) selects the scaffold that becomes a limitation of the independent claim qualifies, because her reasons guided generation and selection at the moment of conception. By contrast, a lab manager who approves purchase orders and triggers runs on a pre-tuned model, but does not frame the problem, curate data in a way that bears on the claimed feature, or exercise claim-defining selection judgment, does not qualify.

Collaborative research is accommodated without dilution. Where several researchers each contribute intellectually at claim-defining junctures, joint inventorship follows; contributors whose value lies in infrastructure, funding, data provision, or high-level oversight fall outside inventorship unless they meet the same threshold at relevant moments. The fact that a model autonomously proposed an intermediate later appearing in the claims does not, without more, disqualify human inventorship; attribution turns on whether human agents directed, interpreted, and selected that output such that conception crystallized, consistent with settled doctrine. Because inventorship carries legal duties, the putative inventor must also be positioned to discharge them. In the US, this includes enablement and the duty of disclosure to the Office (35 U.S.C. § 112(a); 37 C.F.R. § 1.56). In EPO/UK practice, the same materials assist the skilled person with sufficiency (EPC art. 83) and align with expectations of candor and good faith in prosecution. As a matter of international law, the evidentiary practice that implements this doctrine functions as a form-of-disclosure measure compatible with TRIPS Art. 29 and non-discriminatory under Art. 27.¹⁷

For clarity in guidance or legislation, the rule can be stated succinctly:

Augmented inventorship means inventorship in which a natural person exercises meaningful oversight, judgment, or direction over an artificial-intelligence system that contributes to the conception of at least one claimed element; the applicant shall furnish a traceable account demonstrating the person’s role in framing, steering, evaluating, or selecting AI-generated outputs material to the claimed invention.

V. TRACEABILITY IN PRACTICE: AN EVIDENTIARY FRAMEWORK

Administering augmented inventorship requires a modest but structured traceability practice that makes human judgment at the moment of conception legible to examiners and courts without exposing proprietary code or weights. Two concise instruments suffice. A CTR provides a claim-centered map of the generative workflow at junctures material to conception—(i) data provenance (categories and rights basis), (ii) model class and salient parameters at the stages that produced outputs later embodied in the claims, and (iii) an inferential pathway linking inputs/prompts, intermediate outputs, rejection/selection criteria, and external validation to specific claim elements. A HMCS narrates, succinctly and in time order, who did what, when, and why it mattered to conception: problem framing; inclusion/exclusion criteria for data; prompt/threshold design; negative and positive selections; and the human reasons that fixed claim-defining features.

V.A. Ensuring Objectivity and Veracity in CTR and HMCS Disclosures

Because the CTR and HMCS are authored by human applicants, a natural concern is whether such instruments can be relied upon as objective and verifiable evidence of inventorship. Augmented inventorship addresses this concern not by assuming perfect candor, but by embedding CTR and HMCS within existing patent-law mechanisms that already govern truthfulness, accountability, and ex post testing.

First, CTR and HMCS are conceived as contemporaneous records, generated during or immediately following conception-relevant activity, rather than as post hoc litigation narratives. Their probative value derives from their temporal proximity to the inventive act and from their structured, claim-centered format, which constrains opportunistic storytelling by requiring applicants to map specific human decisions to specific claim elements. This mirrors long-standing practices in laboratory notebooks and reduction-to-practice documentation, whose credibility depends less on formal authorship than on internal consistency and contemporaneity.¹⁸ Second, CTR and HMCS operate under the same duty-of-candor and good-faith obligations that already attach to patent prosecution. Material misstatements or omissions regarding human contribution, model configuration, or selection thresholds would expose applicants to inequitable conduct allegations, unenforceability defenses, and potential sanctions.¹⁹ In this respect, augmented inventorship does not introduce a novel honesty problem; it makes an existing one more legible by requiring applicants to disclose the human–machine division of labor that would otherwise remain opaque.

Third, the framework is designed to support proportional ex post verification. In routine prosecution, examiners may rely on the face of CTR and HMCS disclosures, just as they do with enablement statements or declarations. In contested settings—such as priority disputes, inventorship challenges, or infringement litigation—CTR and HMCS can be tested against corroborating evidence, including version histories, prompt logs, experimental records, and third-party validation. Where appropriate, courts may permit protective measures such as redactions, hashed logs, or in camera review to balance evidentiary scrutiny against trade secret protection.²⁰

Taken together, these features position CTR and HMCS not as self-authenticating proof, but as structured evidentiary gateways: they discipline how inventorship claims are made, lower the cost of verification when disputes arise, and align AI-assisted invention with the accountability norms that patent law already enforces.

V.B. Burden, Feasibility, and Proportional Disclosure

The CTR/HMCS pair is content-rich but code-light and proportionate. Applicants provide only what is reasonably necessary to show how human choices steered generative processes at claim-defining junctures; examiners seek deeper artifacts only where AI plausibly shaped conception or material inconsistencies arise. Neither source code nor model weights are required. Properly calibrated, CTR/HMCS reduce back-and-forth during prosecution, sharpen claim construction, and generate a litigation-ready record for enablement/sufficiency. In US practice, the materials sit comfortably with the 2024 USPTO inventorship guidance emphasizing ‘significant human contribution’; in EPO/UK practice, they assist the skilled person with sufficiency (EPC art. 83) and align with EPO Guidelines on AI/ML-based inventions and reproduction of the technical effect.²¹ As a form-of-disclosure measure, the practice is consistent with TRIPS art. 29 and non-discriminatory under art. 27; it adjusts how enablement/sufficiency is shown in AI-intensive projects without changing what must be shown or who may be an inventor.²²

Record-keeping should be contemporaneous and portable. Institutions should maintain internal logs from the start of AI-assisted projects, capturing the minimal set of facts later distilled in CTR/HMCS: objective and constraints; data inclusion/exclusion rationales; architecture class and salient parameter regimes at decision points; thresholding and ranking criteria; and links from intermediate artifacts to the claim language eventually adopted. These are ordinary R&D artifacts reframed for legal salience.

VI. POSITIONING THE DOCTRINE AGAINST EXISTING APPROACHES

The proposed account of augmented inventorship occupies a middle ground between approaches that either render machine generativity invisible or attempt to confer status on non-human systems. Four families dominate. The AI-as-tool view treats models as sophisticated instruments and leaves inventorship analysis unchanged. Its appeal is administrative simplicity; its weakness is evidentiary. By declining to require a structured record of how human judgment interacted with model outputs at the point of conception, it deprives decision-makers of the materials needed to evaluate inventorship and sufficiency. Guidance emphasizing a ‘significant human contribution’ is directionally helpful, but without a claim-element evidentiary grammar it leaves key questions unanswered.²³

At the other extreme is AI-as-inventor—listing the system as inventor when it generated claim-defining features. Descriptive candor meets hard legal limits: across leading forums, current statutes and case law recognize only natural persons as inventors.²⁴ The collateral consequences are substantial: who would execute the declaration, bear duties of disclosure or good faith, assign title, or answer for misrepresentation? Absent comprehensive statutory overhaul, this path is blocked.

Between these poles sit two families that seek to finesse attribution. Owner/proxy attribution would vest inventorship in the controller or proprietor of the system. The attraction is transactional clarity; the problem is doctrinal. Inventorship has long turned on conception, not proprietorship; collapsing the former into the latter abandons that anchor.²⁵ A different strand of disclosure-centered proposals preserves the human-inventor baseline but simply urges applicants to say more about AI involvement. That is sound as far as it goes, and the USPTO’s 2024 inventorship guidance—together with recent analysis in AI-driven drug discovery—underscores the practical need for contemporaneous records.²⁶ Yet the how remains under-specified: most discussions stop at the project level rather than requiring claim-element traceability that shows how human reasons steered model behavior or selection at conception. The result is an information set that is rich yet ill-suited to adjudicating inventorship and sufficiency.

Augmented inventorship improves on these models without disturbing the human-inventor rule; it rejects status for machines, avoids the proprietorship shortcut, and supplies a claim-centered, auditable record calibrated to the doctrinal questions that offices and courts must answer.²⁷ The key innovation is not ‘more disclosure’ in the abstract but traceability aligned to conception: applicants articulate the reasons that directed generative search or selection such that one or more claim elements reflect human judgment. Properly implemented—through concise CTR/HMCS instruments—this preserves trade secrets (no code or weights), reduces error costs, and harmonizes readily across jurisdictions.²⁸ It also engages the emerging literature by converting high-level calls for transparency into a workable evidentiary grammar: where prior proposals and guidance recommend greater candor about AI assistance, this article specifies what must be shown (human steering tied to claims) and how (a proportionate, standardized record).²⁹

VII. ANTICIPATED CRITIQUES AND REPLIES

A first concern is that recognizing AI’s generative role risks smuggling in machine personhood. The doctrine avoids that category error. It preserves the natural-person baseline that leading forums have read into existing instruments and treats models as process evidence only. Inventorship, declarations, assignment, and liability attach to human actors; nothing here alters the holdings that current statutes do not accommodate non-human inventors.³⁰

A second objection is that claim-centered traceability will compel disclosure of trade secrets (source code, weights, proprietary data). That misstates both enablement and the proposal. Enablement (US § 112) and sufficiency (EPC art. 83) require disclosure that enables the invention, not reconstruction of a vendor’s model. The CTR/HMCS are content-rich but code-light: high-level architecture and parameter regimes, data provenance by category and rights basis, and an inferential pathway mapping to claim elements. This supports the skilled person’s ability to practice the invention across its scope—what Regeneron demands in the UK/EPC setting and what US enablement likewise requires—without forcing publication of implementation artifacts.^31,32

A third worry is administrative burden. The proposal is deliberately proportionate. It relies on short, template-driven narratives focused on claim-defining junctures; deeper artifacts (eg, hashed prompt logs) are sought only where AI plausibly shaped conception or where inconsistencies arise. This reduces back-and-forth during prosecution, clarifies claim construction, and supplies a litigation-ready record, reflecting the same direction of travel as recent guidance emphasizing ‘significant human contribution’ while providing the missing evidentiary grammar.³³

A fourth critique targets international coordination. On the contrary, the proposed practice is a form-of-disclosure measure that fits comfortably within TRIPS (Art. 29) and is non-discriminatory under Art. 27: it adjusts how enablement/sufficiency is shown in AI-intensive projects without changing what must be shown or who may be an inventor. The approach is exportable through office guidance and soft-law convergence even before legislatures consider definitional clauses.

A fifth objection is that traceability could chill innovation. The opposite is more likely. By tying inventorship to reasons that steered model behavior at the level of claim elements, the doctrine reduces ex ante error costs (clearer drafting; fewer office actions) and ex post costs (cleaner enablement/sufficiency and inventorship records), while deterring strategic secrecy in precisely those projects where accountability is most needed. Where AI had no material bearing on conception, the records will be brief; where it did, they will already exist as ordinary R&D artifacts reframed for legal salience.

Finally, multi-actor pipelines are addressed without collapsing inventorship into proprietorship. The question is whose reasons fixed which claim elements. Where several researchers each meet that threshold, joint inventorship follows; where actors contribute infrastructure, funding, data provision, or general oversight without shaping conception, they fall outside inventorship but remain compensable through contract or institutional arrangements.³⁴ This preserves the doctrinal link to conception and the UK ‘actual deviser’ inquiry while modernizing the evidentiary showing for AI-intensive work.³⁵

VIII. OPERATIONALIZING AUGMENTED INVENTORSHIP

The doctrine can be implemented incrementally, beginning with administrative practice and only then—if desired—through targeted legislation. Patent offices should issue short CTR/HMCS templates keyed to common AI pipelines (generative chemistry, protein design, materials optimization), with page-limit guidance and a claim-element focus. The CTR/HMCS should do no more than (i) map data provenance, model stage, and salient parameters at junctures material to conception, and (ii) narrate who did what, when, and why it mattered to conception. No source code or model weights are required. This approach aligns with the USPTO’s 2024 inventorship guidance on ‘significant human contribution’ and supports sufficiency under the EPC by making the technical effect reproducible across the scope.³⁶

Proportionality should guide examination. Examiner memoranda should clarify that deeper artifacts (eg, hashed prompt logs or parameter snapshots) are sought only where (i) claims rely on model-dependent behavior or (ii) material inconsistencies appear between the specification and the CTR/HMCS. Otherwise, the default is a concise, claim-centered showing. In the United States, the HMCS interfaces naturally with the inventor’s declaration and the duty of disclosure; misstatements or omissions about AI’s role can be tested against the HMCS without broad discovery into proprietary systems.³⁷ In EPO/UK practice, the same materials assist the skilled person with sufficiency and fit ordinary expectations of candor and good faith in prosecution. As a matter of international law, the practice functions as a form-of-disclosure measure consistent with TRIPS art. 29 and is non-discriminatory under art. 27; it adjusts how enablement/sufficiency is shown in AI-intensive projects without changing what must be shown or who may be an inventor.³⁸

Legislation is optional but straightforward. Jurisdictions that wish to codify the doctrine can add a definitional clause in general definitions (alongside ‘inventor/inventorship’), leaving examination practice to guidance:

Augmented inventorship means inventorship in which a natural person exercises meaningful oversight, judgment, or direction over an artificial-intelligence system that contributes to the conception of at least one claimed element; the applicant shall furnish a traceable account demonstrating the person’s role in framing, steering, evaluating, or selecting AI-generated outputs material to the claimed invention.Pilot programs can road-test the practice with universities and industry consortia. A six- to twelve-month pilot that (i) uses the templates, (ii) measures prosecution frictions (number of office actions; enablement and inventorship objections), and (iii) surveys applicants and examiners on clarity and burden will supply evidence for refinement. If policymakers wish to explore incentive calibration where AI compresses discovery cycles, that can be run as a separate policy pilot—for example, an evidence-based patent-term adjustment tied to CTR showings of acceleration—without entangling it with inventorship doctrine. The core reform here is attribution and traceability; incentives can follow, if needed, on their own record.

IX. CONCLUSION

Patent law can accommodate AI-intensive discovery without abandoning the human-inventor baseline. The doctrine of augmented inventorship keeps inventorship with natural persons while making AI generativity legible precisely where it matters—conception. The immunology analogy supplies the governing intuition: augmentation without displacement. Transposed to invention, that intuition yields directability and traceability, which require human reasons that steer generative processes and a reviewable trail linking those reasons to claim elements.

Doctrinally, the account aligns with existing law: US conception and significant-contribution standards, the UK’s ‘actual deviser’ inquiry, and EPC sufficiency. Administratively, it becomes a proportionate evidentiary grammar—the CTR and HMCS—that is content-rich but code-light, focused on claim-defining junctures, and compatible with forum-specific duties and TRIPS disclosure norms. Practically, this reduces error costs in prosecution and litigation, clarifies claim drafting and construction, and deters strategic opacity in precisely those projects where accountability is most needed.

Crucially, the analysis confirms, and gives a principled basis for maintaining, that applications naming an AI system as inventor (eg, DABUS) cannot, and should not, be granted inventorship. Leading forums have read their governing instruments to require natural-person inventors; nothing in this framework displaces that rule.³⁹ On the contrary, augmented inventorship shows how patent systems can recognize and evaluate machine-assisted generativity without conferring legal status on models: accountability remains human, and patent rights attach only where human judgment fixed claim-defining features and is evidenced accordingly.

If adopted, the doctrine would bring doctrinal coherence and operational clarity to the law of invention in the age of generative systems: augmentation without displacement and accountability without fiction.