When AI generates text, images, music, video, or software, the first legal question appears to be who the author is. One view treats AI as no more than a tool and therefore assigns every output to the user. Another emphasises that AI performs most of the generative work and concludes that the user cannot be the author. A third attempts to treat AI itself as a co-author. Although these answers conflict, they share the same assumption—that every work must have a single source located either in a human being or in AI.

This assumption belongs to the traditional model of creation. On that model, an author first forms an expression in the mind and then fixes it through pen and paper, a camera, a musical instrument, or software. The tool may influence the result, but it is not understood as participating in the generation of expressive structure. Copyright law therefore asks who the human author is, whether the work is original, and how rights move from the author to another party.

That model remains adequate for much traditional creation, so it would be wrong to say that conventional copyright has simply ceased to function. What has lost explanatory force is the image of the author as an isolated origin standing outside the work, together with the presumption that the human being and the tool must occupy mutually exclusive positions. In generative-AI practice, a person may provide only an initial direction while the system produces a large amount of concrete expression. The output then alters the person’s original conception and gives rise to further selection, rejection, and revision. The eventual work is neither an expression already complete in the human mind nor a result formed by AI in the absence of human constraints. Creation has become a recursive structure of feedback.

The proper question is therefore not simply who produced the final words or image. We must ask which expressive differences entered the process, which constraints brought the present work into formation, which human constraints remained effective in the concrete expression, what structural role AI performed, and which subject ultimately incorporated the expression into a relationship they can claim and answer for.

Existing legal approaches are already moving toward this question, although they use different language. The U.S. Copyright Office asks whether expressive elements in an output are controlled by a human author. A Chinese court has examined the natural person’s intellectual input and individualised expression in a particular case. Australian copyright principles continue to connect originality with the identifiable author’s intellectual effort directed toward a particular form of expression. Each approach is trying to determine how a human contribution enters a work. What is missing is a structural theory capable of explaining, within one framework, the creative process, the position of the author, and the separate act of legal recognition.

Beginning with Difference, Constraint, and Sustained Coherence

Sustenesis Theory defines sustenesis as the structural process through which terms of difference, under constraints, form and maintain sustained coherence. Difference is not an obstacle to structure but the condition from which structure can arise. Without difference, there is nothing to select, organise, or sustain. A constraint is any law, rule, boundary, mechanism, norm, feedback process, or procedure that prevents differences from dispersing arbitrarily and makes certain relations effective. Sustained Coherence is the capacity of a structure to retain sufficient identifiability, continuity, and effectiveness across a relevant scale and duration. It does not imply immobility, permanence, or complete material invariance.

These concepts are not labels that can be attached indiscriminately to any subject matter. Sustenesis Theory also requires distinctions between levels. Physical structures, living structures, knowledge structures, subject structures, creative structures, and legal structures can all be analysed through Difference, Constraint, and Sustained Coherence, but each depends on different constraints and produces a different kind of result. The fact that AI can sustain structural knowledge does not mean that it possesses human consciousness or responsibility. The fact that something causes a change in a work does not make it an author. The fact that a creative relationship exists does not by itself require the law to confer an exclusive right.

This distinction between levels sets the limit of the argument. The article does not infer that AI is an author from the fact that AI participates in a structure. Nor does it infer that a person owns the whole output merely because some human constraints entered the work. Creative fact, authorship, and copyright must first be defined separately. Only then can we specify the conditions under which they are connected.

Creation, Expressive Output, and Work

Expression, as used here, is a structure of differences that can be perceived, preserved, and transmitted through language, image, sound, movement, code, or another symbolic medium, and that performs a semantic or formal function within a relevant practice. Expression is neither an abstract idea nor its physical carrier. The same passage may appear in different files, and the same composition may be realised in different performances and recordings. So long as the decisive expressive relations remain identifiable, these instances may refer to the same expressive structure.

Creation is the process of introducing, selecting, transforming, and organising differences in the expressive domain so that, under a set of constraints, they form an identifiable and relatively sustained expressive structure. This definition describes structural formation before deciding which participants, if any, qualify as authors. A human being may create; a system composed of humans and tools may carry a creative process; and AI may generate and transform expressive differences without subject consciousness. To avoid conflating structural production with the committed act of a subject, this article uses generation for the process through which AI produces candidate content under the constraints of data, parameters, context, prompts, tools, and feedback. Generation may be part of creation without becoming, for that reason alone, an act of authorial commitment.

An expressive output is the perceptible result formed by a creative or generative process at a particular moment. It may be only a candidate output, or it may already have the stability of a work. A work is an expressive structure capable of maintaining a distinguishable identity under a particular medium, version, timescale, or generative rule. A work is not an absolutely finished and permanently closed object. An article may acquire the identity of a work in a published version and still be revised later. A piece of generative music may continually produce new sounds while retaining its identity through a sustained generative rule. Interactive art may preserve its decisive structure even while participants continually alter its content.

What makes a work possible is therefore stage-specific identifiable stability, not the final closure of creation. Stage-specific identifiable stability means that, at a particular point of judgment and on the relevant scale, an expressive structure can be distinguished from other possible results and can be preserved, reproduced, retrieved, or generated further under relatively stable rules. The work’s boundary may change across later versions, circulation, and reinterpretation. That openness does not cancel its identity at the present scale.

A work in the sustenetic sense must also be distinguished from a work in the legal sense. The former is an expressive structure that has reached stage-specific stability of identity. The latter is an object recognised by law under the jurisdiction’s requirements of originality, expression, and other statutory conditions. Not every identifiable expressive structure needs copyright protection. Facts, standard forms, mechanical results, and AI outputs lacking the necessary relation of human authorship may all be stable expressions without becoming legally protected works.

The Creative Sustenetic Structure Formed by Humans, AI, and Expressive Outputs

At the factual level, AI-assisted creation first appears as a structure jointly formed by human participants, AI systems, expressive materials, candidate outputs, and feedback. At this stage the human participant cannot yet be called the author, and not every output can yet be called a work, because authorship and work-status are conclusions still to be established.

Human participants may introduce themes, purposes, source materials, formal requirements, value orientations, and aesthetic preferences. AI systems generate, transform, and reorganise expressive differences under the constraints of training data, model architecture, parameters, context, prompts, tools, and feedback. Candidate outputs are not passive endpoints. Once presented, they become new differences and new feedback. They alter human understanding and cause an original direction to be refined, corrected, or rebuilt. The actual process is therefore not a straight line in which a human gives an instruction and AI executes it. It is a recursive interaction among human constraints, the generative structure of AI, and feedback from results.

Whether AI belongs to a particular creative structure cannot be decided merely by asking whether AI was used. If a person uses AI only to count words or convert a file format, or if the intervention has no actual effect on human choices or final expression, AI is present in the workflow but is not a constitutive factor in that act of creation. AI becomes a genuine constitutive factor only when the expressive differences it introduces, or a reorganisation of constraints triggered by its feedback, retain an actual formative effect on the selection, organisation, or revision of the stage-stable result. Even content later deleted may have participated in creation by changing the person’s understanding and subsequent decisions. Constitutive factor here means only that AI performed a structurally non-negligible role. It does not mean that AI is a subject, author, or right-holder.

For the same reason, model developers and the authors of training materials do not automatically become co-authors of every output merely because their work supplied background conditions for the model’s capacity. Background constraints must be distinguished from constitutive constraints. Background constraints define how a class of results can be produced. They include model architecture, training methods, linguistic rules, and general tool functions. Constitutive constraints operate within a particular creative process. They direct the selection, exclusion, transformation, or organisation of candidate differences and retain an actual effect on the selection, organisation, or revision embodied in the stage-stable result. A constitutive effect is traceable through the actual process of formation. It does not require the person’s input to survive word for word in the output, nor does it require proof that regenerating the output would necessarily produce different surface content. Background constraints may give rise to separate issues concerning contracts, training data, reproduction, or infringement, but they do not automatically establish authorship of each particular output.

This distinction cannot be used to erase the rights of prior authors. If model output reproduces or substantially retains protected expression, the dispute concerns the relationship between the source structure and the output structure. It cannot be displaced merely by establishing authorship on the part of the final user. Training input, model formation, particular acts of generation, and distribution of output are connected but distinct structural levels. A rights question at one level cannot be absorbed into another.

Authorship as a Subject-Position Within the Creative Relation

Sustenesis Theory does not regard the subject as the original source of all structure. This article follows the theory’s existing definition—a subject is a higher-level sustenetic structure capable of consciousness, reflection, commitment, responsibility, and participation in a shared world. AI may preserve, retrieve, and operate structural knowledge, and its outputs may have semantic effect. Structural knowledge is not subjecthood, however, and generative capacity is not authorship.

Authorship is neither an isolated inner property of a human being nor an undifferentiated property shared by an entire human–AI system. In this article, authorship is a sustenetic concept rather than a legal test already adopted uniformly across jurisdictions. It is defined as a structural position within a creative relation that can be attributed to a subject. A subject occupies this position when, through directional constraints, judgment, selection, verification, and adoption that have an actual formative effect, the subject brings expressive differences into a stage-stable result and incorporates that result into an expressive relationship they can claim and answer for.

Attribution here first concerns authorship at the factual level. The relevant constitutive constraints must be traceable through the creative process to the formation, maintenance, or revision carried by the same subject, and the result must be adopted by that subject. This is not merely a relation of physical causation. Nor does it mean that the law has already recognised the subject as an author or conferred copyright.

A shared world is the social sustenetic structure within which expression, commitment, and responsibility can be recognised, answered, and revised by multiple subjects. An expressive commitment arises when a subject incorporates an expression into a relationship in which the subject can claim, explain, and answer for it. Publication is not required, and the commitment is not a guarantee that the content is true. An unpublished work may still have authorship if the subject already occupies this position of claim and answerability within the creative relation.

Intention, in this account, is not an unverifiable private wish inside the mind. It is a directional constraint maintained by a subject toward a future expression. Judgment occurs when the subject selects, rejects, or revises among possible results under higher-level constraints of meaning, form, fact, value, or purpose. Verification compares a stage-specific result with those governing constraints and converts identified deviations into a further round of adjustment. It may concern factual accuracy, but it may also concern argumentative structure, character relations, musical direction, visual composition, or expressive style. Adoption occurs when a subject incorporates the result, at the relevant stage, into an expressive relationship they can claim and answer for. Responsibility does not mean unlimited liability for every technical consequence. It means that the subject can answer for and explain the adoption, and can bear those consequences that the legal system assigns in light of the subject’s control and conduct. Intention, judgment, selection, verification, and adoption are functional dimensions of authorship, not five mandatory steps that every creative process must display in sequence.

Authorship cannot be established merely by showing that a person remained involved. The person’s constraints must have a constitutive effect on concrete expression. A standing instruction to use Chinese, or a continuing prohibition against certain words, may be real but cannot by itself explain why the work took its present expressive form. Time spent, number of prompts, number of generations, and money paid cannot substitute for authorship. Human constraints must exert an identifiable formative effect on the expression. They must actually restrict the acceptable expressive range or enter the relations of selection, organisation, and revision. Evidence that removing or replacing a constraint would alter the result can support a finding of formative effect, but it is not the only necessary test. Randomness in generative systems and overlapping contributions by multiple participants make a strict single-counterfactual test unsuitable as a universal standard.

Creative constraints usually form a network rather than a straight line. Intention, source material, prompts, candidate outputs, selection, revision, and feedback may be preserved, transformed, weakened, interrupted, and rebuilt at different stages. The path through which these constraints ultimately retain an actual effect on the expressive result may be called a creative constraint chain. The concept does not require a unique human cause for every word or element. It asks whether the person’s governing constraints genuinely passed through the generative process and entered the identifiable expressive relations of the work.

On this basis, entering one broad prompt and accepting the first result ordinarily proves only that a person triggered generation. Producing many outputs and choosing one arbitrarily may remain no more than an acceptance of chance. By contrast, when a person supplies source material with expressive content, continually adjusts structural relations, rejects and revises candidate results in a directed manner, and integrates multiple rounds into an identifiable whole, authorship may form within that relationship. Selection and arrangement may also constitute authorship at a higher level even when the individual AI-generated elements do not themselves contain human authorship. The scope of authorship must correspond to the expressive scope actually formed by the person’s constitutive constraints. A local contribution cannot automatically expand to cover the entire output.

Final approval alone cannot retrospectively turn into human creation a result that had no constitutive relation to the approving subject’s prior constraints. Approval contributes to authorship only when it carries forward a real relationship of directional constraint and judgment. Otherwise, anyone could appropriate an independently existing output merely by declaring it to be their work, and authorship would collapse into possession after the event.

Originality is a legal eligibility constraint through which each jurisdiction identifies protectable expression. Its precise meaning belongs to the applicable law. This article does not invent a universal category of “sustenetic originality.” What Sustenesis Theory reconstructs are the structural facts on which legal originality may depend—whether the expression mechanically reproduces an existing concrete expression, which human constitutive constraints entered which parts of the output, and whether those relations can be traced through evidence. Whether the statutory threshold of originality is met remains a question for the legal structure of the relevant jurisdiction.

The preceding analysis explains how authorship can arise within a creative relation. It does not yet show why the law must confer copyright. Creative structure belongs to the factual level, authorship to the level of attribution to a subject, and copyright to the social and legal level. They are connected, but they are not the same structure.

Sustenesis Theory understands law as an institutional structure sustained through rules, institutions, procedures, interpretation, enforcement, social recognition, and relations of responsibility. Law is not simply a command written in a text. It persists as law only when rules can be identified, invoked, interpreted, and enforced through procedures, and when they create relatively stable expectations of conduct among social participants.

On that basis, copyright can be defined as an institutional structure through which the legal sustenetic structure makes a sustained and enforceable configuration of recognised expressive objects, authorial identity, initial ownership, subsequent transfer, boundaries of use, limitations, and remedies. Copyright is not a natural property inside a work, nor is it ownership formed by the automatic outward extension of authorship. Authorship is an important input to legal judgment, but the law must still decide which expressions qualify as protected objects, which parties can hold rights, how far protection extends, when others may use the work, and what remedies are available when conflicts arise.

An author and a right-holder are not the same. The author in this article is first the subject who occupies the position of authorship within a creative relation in the sustenetic sense. This does not necessarily coincide with the legal author created by a jurisdiction’s institutional rules. Some jurisdictions may treat an employer, commissioning party, corporation, or other organisation as the legal author in specified circumstances. Other jurisdictions preserve the human creator’s authorial identity while assigning initial economic rights to an employer. These rules are legal configurations of status and entitlement. They do not prove that an organisation possesses creative consciousness in the sense used by Sustenesis Theory.

A right-holder is a legal position configured through rules of creation, employment, assignment, inheritance, or another recognised route, enabling a person or organisation to assert and exercise particular copyrights. A corporation may be a right-holder, and a natural-person author may have transferred some economic rights. Legal right-holder status is an institutionally sustained position from which claims can be made. It must not be confused with the sustenetic concept of a subject capable of consciousness, reflection, and responsibility.

Legal recognition is the process through which rules, procedures, adjudication, and enforcement stabilise a claimed creative relationship as a legal identity that can be invoked and implemented within the institution. What it produces is rule truth—a judgment that holds and can be enforced under a particular set of legal rules, evidentiary limits, and procedures. It is not automatic proof of creative fact, scientific truth, or moral truth. Genuine authorship may not yet have been recognised by law, and an effective legal determination of authorship may reveal only part of the actual creative structure because of the evidence and institutional rules available to the court. That is why a judgment can have legal force while remaining open to theoretical analysis and institutional revision.

A more accurate account of AI-assisted creation is therefore not that the human, AI, and work are themselves “copyright.” Human participants, AI systems, expressive materials, and outputs first form a creative sustenetic structure. Authorship arises within that structure as a position attributable to a subject. Copyright law then applies legal-level constraints to recognise some instances of authorship, constitute some expressions as legal works, and configure rights, limitations, and responsibilities.

Why Should the Law Protect Authorship?

Even where authorship can be identified as a fact, the law could choose not to establish copyright, or could protect creation by entirely different means. Difference, Constraint, and Sustained Coherence explain how structures form. They cannot, on their own, generate the normative conclusion that the law ought to protect them. Moving from creative structure to copyright therefore requires an account of the higher-level structure that the institution is meant to sustain.

This article introduces a normative premise explicitly and separately from the descriptive analysis provided by the three elements. Given human sociality and interdependence, law ought to preserve the conditions under which subjects can continue to express themselves, answer for their expressions, and participate in shared life. At the same time, it ought to preserve the public conditions under which knowledge and culture can continue to be generated and circulated. The coordination of these conditions across relations among subjects is called higher-level social sustenesis.

The account of value used here also follows the existing definition in Sustenesis Theory. Value is the directionality displayed when a sustenetic structure maintains, repairs, extends, or enhances itself. Stability alone is not value. Directionality must always specify the structure, scale, and level toward which it is directed. Under the normative premise introduced here, the test is not whether a particular rights structure can perpetuate itself indefinitely. The test is whether the conditions of subject expression, circulation of knowledge, and generation of shared culture can be coordinated across relations among subjects.

“Higher-level” refers only to a wider relational scope extending from a single creator and a single work to other creators, users, systems of knowledge, and shared culture. It does not mean that the whole automatically has greater value, and it does not permit subjects to be sacrificed arbitrarily in the name of the whole. Social sustenesis is not the self-perpetuation of whatever order already exists, nor does institutional stability establish legitimacy. Sustenesis Theory distinguishes stability from value. Suppression, monopoly, and falsehood can all form stable structures.

Creation is not a private event sealed inside an isolated individual. Language, music, image, and thought all depend on expressive conditions formed within a shared culture. Yet individual subjects must also be able to establish their own expressive direction within those conditions. Their expressions must be capable of entering a shared world under an identifiable attribution and, where appropriate, of supporting a relatively stable return. If anyone’s work can be copied without limit, stripped of attribution, and substituted by others, the social conditions necessary for sustained creative commitment may erode. By recognising authorship and providing bounded relations of control, attribution, and return, law can help maintain the subject’s capacity to continue creating and to answer for what they express. It can also preserve provenance, paths of responsibility, and cultural memory.

The same normative premise, however, requires knowledge to circulate, be learned, criticised, transformed, and incorporated into later creation. If copyright expands without limit and permanently encloses cultural material in the hands of a small number of right-holders, it damages the public conditions of shared creation. Copyright does not derive its legitimacy from the absolute control of the individual, or from the mere stability and persistence of the institution. Its justification depends on whether it preserves both subject expression and attributable responsibility while leaving viable conditions for the circulation of knowledge, public use, and subsequent creation. Limited terms, fair use or fair dealing, the public domain, statutory licences, and educational and research exceptions are ways in which law rebalances constraints at this level.

A limited institutional conclusion follows. When evidence shows that recognising a genuine relation of authorship would help sustain these shared conditions without disproportionately obstructing the circulation of knowledge or later creation, the law has a defensible reason to recognise it. The scope of authorial attribution should not extend beyond the authorship actually formed. The scope of particular rights must still be determined by other constraints at the legal level.

This principle does not itself settle how long protection should last, which exceptions a jurisdiction should adopt, how platform responsibility should be allocated, or how damages should be calculated. Those choices remain constrained by enforceability, evidentiary cost, technological conditions, competing rights, and the public interest. Sustenesis Theory supplies a direction of judgment and a structural boundary. It does not bypass legislation and adjudication to generate a complete legal code.

The current registration and policy position of the U.S. Copyright Office does not exclude copyright merely because AI was used. It asks, case by case, which expressive elements can be attributed to a human author. In its 2025 report Copyright and Artificial Intelligence, Part 2: Copyrightability, the Office concluded that, given currently available generative technologies, prompts ordinarily do not provide sufficient control to make the user the author of the resulting expression. Human expression that remains perceptible in the output, however, may be protected, as may creative selection and arrangement and post-generation modification that satisfies originality. Purely AI-generated material, or material over whose expressive elements a human did not exercise sufficient control, remains unprotected. The Office therefore tends to treat hybrid outputs as divisible, protecting the human-attributable contribution rather than assigning the whole AI output to the user. In Thaler v. Perlmutter, the U.S. Court of Appeals for the District of Columbia Circuit held in 2025 that a work must first be authored by a human being to qualify under the Copyright Act of 1976. The case addressed only the narrow claim that AI was the sole author. It did not determine the threshold for human contribution in the many forms of hybrid human–AI creation. U.S. Copyright Office report; Thaler decision

In one effective first-instance decision, the Beijing Internet Court followed a different analytical route. In Civil Case (2023) Jing 0491 Min Chu No. 11279, the court evaluated the disputed image as a whole and reconstructed the process that produced it, including model choice, the design and ordering of prompts, parameter settings, continuous adjustments, and final selection. On the facts of that case, the court found that these activities embodied the natural person’s intellectual input, aesthetic choice, and individualised expression. It therefore held that the image as a whole was a work of fine art and that the plaintiff, who had used the AI tool, was its author and copyright owner. The court also stated that, under current Chinese copyright law, the AI model lacked free will and legal personality and therefore could not be the author. Neither party appealed, and the first-instance judgment took effect. The decision is an important reference point but should not be presented as a nationally uniform Chinese rule governing all AI-generated material. Beijing Internet Court judgment; statement confirming that the judgment took effect

For literary, dramatic, musical, and artistic works governed by Part III of Australia’s Copyright Act 1968, the existing framework continues to depend on an original work, an author, and rules of ownership. It has not established a special statutory author for generative-AI outputs. In IceTV, a compilation case unrelated to AI, the High Court of Australia stated that an original work must originate in the author’s independent intellectual effort. When assessing the quality and substantiality of what was copied, the relevant skill and labour must be directed to the particular form of expression rather than merely to information or commercial arrangement. The case did not establish a copyright standard for AI output. At the time of writing, the Australian Government continues to consider greater legal certainty for AI-generated material through its Copyright and Artificial Intelligence Reference Group. Australia therefore cannot yet be described as having adopted either the U.S. Copyright Office’s divisible-elements approach or the whole-output approach taken in the Beijing case. Australian Copyright Act 1968; IceTV decision; Australian Copyright and Artificial Intelligence Reference Group

These differences show that the central dispute is not only whether AI can be an author. The United States emphasises human control over expressive elements. The Beijing case emphasised intellectual input and individualised expression across the generative process. Australia continues to rely on the traditional relation between an identifiable author and the particular expression in which originality is claimed. All three approaches face the same underlying difficulty. When human intention can become concrete expression only through a generative model, which human constraints have remained effective in the work, and which merely triggered, influenced, or selected a machine-produced result?

Sustenesis Theory does not answer by calculating a ratio between human labour and AI generation. The percentage of words contained in prompts, the volume of text generated by AI, or the amount changed in post-editing cannot by themselves determine authorship. The inquiry must instead identify the sources of expressive difference, determine which constraints exerted a constitutive effect on concrete expression, examine whether those constraints were maintained and revised through feedback, establish whether the result reached stage-specific identifiable stability, and identify the subject who incorporated it into an expressive relationship they can claim and answer for.

The sustenetic analysis of authorship in AI-assisted creation can be organised around several connected inquiries.

The first is to identify the terms of difference at the factual level. These include source text, images, sounds, and conceptions supplied by humans; candidate content generated by AI; capacities of transformation supplied by models and tools; changes between successive outputs; and expressive elements retained or excluded in the final result. Without reconstructing the actual process, abstract references to “human contribution” or “AI contribution” have no determinate object.

The second is to distinguish background constraints from constitutive constraints. A model, platform, and general prompting practice define a space of possibilities. Authorship requires human constraints to have an actual effect on the concrete expressive relations of the present work. Requirements concerning theme, style, or language may be too general to do more than define a category. Repeatedly pressing a generate button does not increase authorship by itself. The law should ask where the person actually excluded alternatives, changed expressive relations, and caused those changes to remain effective in the result.

The third is to determine whether the constraints formed a position of authorship attributable to a subject. Did the person’s directional constraints pass through the generative process? Did judgment and verification continue to operate around a sufficiently stable expressive purpose? Did final adoption carry forward a genuine prior creative relation? Did the subject incorporate the result into an expressive structure they can claim and answer for? Actual publication is not necessary for authorship. Nor must an author foresee every detail. Photography, improvisation, and chance-based art also contain unforeseeable differences. The decisive question is whether, after contingent differences appeared, the subject organised them into the work through constraints that had a formative effect.

The fourth is to determine the level at which the work reached stage-specific stability. For an image, the relevant object may be a fixed visual form. In long-form human–AI writing, it may be the text integrated through many rounds of generation and editing. In interactive or generative art, the work’s identity may be maintained mainly through its rules, interface, and boundaries of variation. Different kinds of works cannot be judged through a single mechanical scale of control.

Only then should analysis enter the legal level. The law must decide which expressive scope qualifies for protection, how author and right-holder are to be distinguished, whether the pre-existing rights of others are affected, and whether protection is consistent with higher-level social sustenesis. Because legal recognition depends on evidence, prompts, parameters, versions, source materials, generation histories, selection records, and human revisions may all help establish a creative constraint chain. Their purpose is to make an existing relation visible to the institution. A large quantity of records cannot manufacture authorship where the relation did not exist.

This standard will not remove every difference between cases, but it converts the vague question of whether human contribution was “enough” into a set of analysable structural questions. It also avoids two opposite errors. One assigns the entire output to a person whenever they press a button. The other denies human authorship whenever AI produces concrete words or images. Sustenesis Theory recognises that generative capacity may be distributed across a human–AI system while preserving authorship as a structural position attributable to a subject.

Rights and Responsibilities Require Consistent Grounds of Attribution

Rights and responsibilities need not be fully symmetrical in subject, scope, or degree. An author may transfer economic rights. Employers, platforms, developers, and users may bear different responsibilities under different legal relations. A minor may hold rights without bearing the same legal consequences as an adult. Sustenesis Theory therefore cannot infer that whoever holds rights must bear all responsibility.

What must remain consistent is the treatment of the same fact of control within the same case, although that fact may have different legal significance under authorship rules and liability rules. If a person relies on final adoption as part of a claim to authorship, they cannot simultaneously deny that they adopted the result. Final adoption alone, however, does not prove knowledge of infringement, breach of a duty of care, causation of harm, or liability for every consequence.

If a platform or developer did not enter the constitutive constraints of a particular work, this excludes only the attribution of responsibility based on authorship of that work. It does not exclude responsibility that may arise from model design, training conduct, notice obligations, contract, or other legal rules. Each party’s responsibility must be assessed within the legal relation to which it belongs.

The requirement is therefore not quantitative symmetry between rights and responsibilities, but consistency between factual findings and grounds of attribution. Responsibility must still be configured according to each party’s actual control, state of knowledge, duties of care, contractual relations, and causal connection to harm. This section states a sustenetic demand for internal institutional coherence, not a liability rule already shared by existing legal systems. It prevents a legal institution from relying on human control to establish authorship while denying that same control at the factual level when responsibility is considered. It also prevents the structural role of AI from being misdescribed as an independent capacity to bear institutional responsibility.

Theoretical Boundaries and Future Questions

The conclusions of this article depend on the distinction Sustenesis Theory presently draws between AI and a subject. AI can sustain structural knowledge and may exhibit forms of structural understanding, but those capacities do not by themselves establish consciousness, reflection, commitment, responsibility, or participation in a shared world. In the absence of sufficient grounds for concluding that current AI satisfies those conditions, this article treats it as a non-subject constitutive factor within a creative structure. The quality of an output cannot by itself alter that classification.

If a future AI were to form a subject structure capable of sustaining its own identity, reflecting on its actions, making commitments, bearing responsibility, and participating in a shared world, the question would change. What would then require reconsideration is whether AI had become a subject and whether the social and legal structure should recognise it as a legal right-holder. The answer should not be inferred merely from whether its work resembles human creation. Creative capacity, subject status, and legal eligibility for rights remain three distinct levels.

Sustenesis Theory also cannot, by itself, resolve every question concerning the legality of training data, style imitation, market substitution, platform immunity, transnational jurisdiction, or damages. These issues concern different objects, subjects, and institutional levels. Each requires its own reconstruction of differences and constraints. Gathering all of them under the heading of “AI copyright” would obscure the problem once again.

What can be established is that the traditional model of the isolated author no longer adequately describes AI-assisted creation, while the human author has not disappeared. The author moves from being the sole and immediate cause outside the work to being the subject-position that establishes direction, sustains constitutive constraints, exercises judgment, and undertakes expressive commitment within a creative relation. AI’s structural role is recognised without inventing a legal subject that has not yet been established. A work is no longer understood as a static object closed once and for all, but as an expressive structure that has reached stage-specific identifiable stability under a particular medium, version, rule, and timescale.

Conclusion

Copyright disputes in the age of AI appear to ask whether a work belongs to a human being or a machine. What they actually reveal is the excessive simplification of the creative process in traditional copyright theory. Creation has never moved without mediation from a closed subject into a passive work. Under traditional tools, however, the complexity of the intervening relations could usually be compressed into a binary structure of author and work. Generative AI has made those previously compressed relations visible.

Sustenesis Theory first reconstructs the process as a creative sustenetic structure formed by human participants, AI systems, expressive materials, candidate outputs, and feedback constraints. The expressive differences introduced by AI, or a reorganisation of constraints triggered by AI feedback, make AI a genuine constitutive factor in a particular creative process only when they retain an actual formative effect on the selection, organisation, or revision of the stage-stable result. Structural participation does not confer subjecthood.

On that factual structure, authorship is defined as a position within the creative relation that can be attributed to a subject. The author is not the sole physical cause of every expressive difference. The author is the subject whose directional constraints, judgment, selection, verification, and adoption exert an actual effect on concrete expression, and who incorporates the result into an expressive relationship they can claim and answer for. Human authorship is determined neither by the number of prompts, the time spent, nor final possession, but by whether human constraints entered the work with a constitutive effect.

A work is an expressive structure that has reached stage-specific identifiable stability at a relevant scale, not the absolute closure of the creative process. Copyright belongs to a higher legal sustenetic structure. It is the sustained configuration of authorship, protected works, ownership, boundaries of use, limitations, and remedies. Creative fact, authorship, and copyright are thereby distinguished and reconnected.

The normative basis for legal protection is not that every stable creative relation should become exclusive property. The normative premise introduced in this article requires law to preserve both subject expression and attributable responsibility while leaving viable conditions for the circulation of knowledge, public use, cultural memory, and later creation. Copyright is justified only insofar as its configuration can withstand the test of higher-level social sustenesis. Its legitimacy does not arise from structural stability or persistence alone.

What must be redefined in the age of AI is therefore not whether a machine can possess a human copyright, but how the law should identify authorship genuinely formed within a distributed creative process. The central claim can be stated as follows:

AI-assisted creation is a multi-level creative sustenetic structure formed by human participants, AI systems, expressive materials, and outputs. Authorship is the structural position within that relation that can be attributed to a subject. Copyright is the institutional structure through which law configures authorial identity, ownership, boundaries of use, limitations, and remedies around recognised expressive objects. Whether that configuration is justified must be tested, under the normative premise introduced in this article, by its contribution to higher-level social sustenesis.

This reconstruction neither reduces AI to an irrelevant passive tool nor prematurely elevates it to legal subjecthood. It recognises that creation has become a structural process within human–AI relations while preserving the subject-based foundation on which authorship, responsibility, and rights can be established.

References