As AI churns out increasingly convincing images, music, and prose, a chorus of voices from Big Tech and their academic followers claim machines should be recognized as authors. Caterina Moruzzi pushes back, arguing that authorship is not about output alone, but about agency, responsibility, and expression—qualities today’s AI systems do not possess.

In 2018, the computer scientist Stephen Thaler applied to register the image titled A Recent Entrance to Paradise with the U.S. Copyright Office. On the application form he listed a generative system he created, the “Creativity Machine,” as the work’s sole author. After several requests for reconsideration, the United States Court of Appeals for the District of Columbia Circuit issued its final decision on March 18, 2025. The court framed the issue in simple terms: “Can a non-human machine be an author under the Copyright Act of 1976?” Its answer was no: “The Creativity Machine cannot be recognized author of a copyrighted work because the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.”

What may look like a purely legal ruling has become a landmark case for AI and creativity. By reaffirming human authorship as “a bedrock requirement” for copyright, the court traced a clear demarcation around humans as the sole possible locus of authorship, taking a stand on the broader question of what authorship really is.

Not everyone accepts the human-only requirement. Law professor Ryan Abbott argues for “AI legal neutrality”, which holds that law ought not discriminate between human and machine behavior when they perform the same creative or inventive activity. Computer scientist Advait Sarkar pushes from the cultural side: authorship is historically and culturally contingent and personhood has not always been a prerequisite for attribution of authorship, so our concepts should evolve to encompass emergent AI-mediated practices.

The rest of this piece argues for keeping authorship human while updating it beyond the solitary genius: today’s AI lacks the capacities authorship presupposes, but this should not prevent us from recognizing and crediting the distributed human practices through which much creation happens.

What is an author?

On the surface, authorship might seem to be nothing more than a matter of causal production: whoever, or whatever, is the most direct cause of a work’s existence would be its author. But as soon as we look at familiar cases from everyday life and art, it becomes clear that our account of what an author is asks for more than mere causal production.

___

We react very differently to cases where the output looks similar, but the process behind it differs.

___

Take the following example: I accidentally hit a few pots of paint on a warehouse shelf, the paint spills, and it leaves splashes on the floor. By contrast, consider the work Autumn Rhythm (Number 30). Both are just pigment on a surface, and they might even be visually similar, but while in the first case we would not be tempted to debate whether I am the “author” of the paint splash, we clearly recognize Jackson Pollock as the author of Autumn Rhythm.

Now, compare a toddler striking piano keys at random with an experimental music piece by Pauline Oliveros. In both cases, we might hear unexpected clusters of sound, even long stretches of silence. The toddler caused the sounds, but causation alone does not seem enough for an authorship claim.

Finally, think of the patterns carved by wind on a rock, and a land sculpture by Andy Goldsworthy. Both may involve similar materials and shapes, and yet we do not credit the wind as their “author.”

SUGGESTED VIEWING AI and the future of film with Martha Fiennes With Martha Fiennes

We react very differently to cases where the output looks similar, but the process behind it differs. Our judgment distinguishes between causation and authorship: we do not credit the wind, the toddler, or me spilling paint on the floor by accident, despite the fact that they are causally necessary. We do instead credit Goldsworthy, Oliveros, Pollock, and countless others, although they outsource part of the labour to materials, instruments, assistants, and even partly to chance.

The idea that authorship requires more than mere causal production has been developed across philosophy and copyright scholarship. Although these fields do not agree on a single definition of “author,” they tend to converge on three dimensions: agency, responsibility, and expression.

Agency

We consider a work “authored” when it is the result of an agent’s intentional activity.

The philosopher Christy Mag Uidhir analyses authorship explicitly in terms of an agent’s intention-directed activities: “x is the author of w as an F” only if w is the product of an intention-directed activity for which x is the source. Similarly, in copyright law, attribution of authorship follows rules of agency and intentionality: in the 2010s, the monkey Naruto became famous by triggering a camera that was left on a tripod in the middle of the Indonesian forest by the photographer David Slater. Although being the most direct cause of the resulting selfie, Naruto is not deemed the “author” of the photograph, because the act lacked the kind of intentional agency that authorship presupposes.

Responsibility

Authorship is not only about who did something; it is also about who can be held to account for what has been done.

In his lecture “What is an Author?” (1969) Michel Foucault suggests that the “author function” emerged when texts became subject to censorship and punishment. Authorship can be understood as a normative status that connects a person to a work in terms of credit, accountability, and liability: to be the author is to be the one to whom praise and blame attach in relation to that work. In copyright, too, authorship and responsibility are tightly connected: the law assumes that an author is a being who can understand and comply with duties, be held accountable, and enter contracts.

Expression

A third dimension of authorship is about a work’s connection to the author’s way of seeing and being in the world.

Expression explains why we care who the author of a work is, as knowing its authorship shapes how we engage with the work and how we situate it within broader social and cultural contexts. Hegelian and Kantian theories of personality feed directly into modern doctrines of moral rights, which treat works as outward expressions of the creator’s self. The rights of attribution and integrity make sense only on the assumption that a work is bound up with an author’s identity, and that distorting it is a kind of personal offense.

___

Current AI systems do not have inner lives of points of view that can be expressed in their output.

___

The picture that emerges is very different from the simple causal view. Authorship is a socially and institutionally constructed status, organized around a cluster of features that explain why our authorship judgements are sensitive not only to outputs but to the agency, responsibility, and expression involved in creating them.

Is AI an author?

If authorship were only a matter of causal production, then the answer would be straightforward: if an AI model produces an image or a text—if it causes the image or the text to exist—then it can be considered an author. This sits well with the way we talk about AI in everyday language: we say that the model “wrote” a poem or “generated” an image. However, as we have discussed, the causal story is only a starting point. The more difficult question is whether AI systems can occupy the status function that we reserve for authors.

Agency

Current AI systems do not meet the standard of agency and intentionality that is required for authorship. They are statistical engines that produce outputs when they are prompted or otherwise activated. Their “choices” are the result of optimizing next-token probabilities or denoising algorithms. The unpredictability of AI outputs that is sometimes put forward as evidence of autonomy and creativity of the models is not the same as intentionality. Even when there is chance or indeterminacy in the process of human creation, as with Pollock’s drip paintings, these aleatory elements are part of a higher-level intentional conception of the creator. In the case of AI, the intention can be traced back to the human designers who are responsible for the model’s architecture and the human users who prompt and curate the outputs. 

Responsibility

AI systems cannot be sued, they have no property or interests to lose, they do not experience moral concerns, and nor do they revise their works in light of criticism. When an AI model “hallucinates” damaging statements, courts look beyond the model to the human actors and organizations behind it: to the company that released the model without adequate safeguards, the engineers who built the model, or the users who republished the statements without fact-checking them. The challenge of identifying whose, exactly, the responsibility is for a harm caused by AI gives rise to a responsibility gap. What is widely agreed, still, is that the AI system cannot be considered to have responsibility, as it lacks moral agency.

Expression

Current AI systems do not have inner lives or points of view that can be expressed in their output. We can talk loosely about “the style of Midjourney” or “ChatGPT’s voice,” but what we are really tracking are the artefacts of design choices by OpenAI or other companies, filtered through vast corpora of human-made material. Whatever “style” appears in their outputs is the result of training data and prompting supplied by humans.

Rethinking authorship

When the D.C. Circuit in Thaler v. Perlmutter insists on the fact that “machines are tools, not authors” it does so in accordance with the basic philosophical insight that authorship is grounded in agency, responsibility, and expression that current AI systems do not possess.

At the same time, the court’s decision risks reinforcing the narrow notion of the author as an individual, which lies at the normative heart of our vision of copyright.

Authorship is not a matter of an isolated individual acting alone, but is instead distributed across collaborative processes, mediated by tools, and structured by institutions that decide whose names appear on the work. This is true also of processes that do not involve the use of AI, but AI magnifies the tension between individual and collective views of authorship, revealing the limits of a legal and cultural apparatus built around an image of the individual author that does not fit the reality of distributed creation.

So, what should we do?

1. Keep authorship human

We should hold on to the basic principle that authorship should remain a human status. AI systems are powerful tools that reshape the landscape in which human authorship is exercised, but they are not, themselves, the kinds of entities that our existing theories and practices can recognise as “authors.” Attributing authorship to machines would not only depart from a philosophical understanding of what an author is but also make poor legal sense. Copyright and patent rights are meant to work as an incentive mechanism; a system that cannot own property, cannot be sued, and cannot care about its reputation gains nothing by being named as “author.” But in order to resist AI authorship we do not need to cling to a picture of an individual human author as an unassisted genius. Human authorship can remain central even in technologically dense environments, where creation takes place through networks of people and tools.

2.  Improve credit and transparency mechanisms

Institutions that mediate culture, like publishers, galleries, and media platforms, can normalize transparent crediting of AI assistance. This labelling respects audiences’ interest in knowing how works were made, demystifies the role of AI tools, and gives human authors an opportunity to take ownership of the way they integrate these tools into their practice, rather than hiding them out of fear that disclosure will invalidate their claims to authorship. At the same time, we should invest in technical infrastructures, like content credentials that make it easier for human creators to track and demonstrate their contributions in AI-assisted workflows.

3. Locate the threshold of human authorship in AI-assisted works

In March 2023, the US Copyright Office issued a policy statement on works containing AI-generated material, which states that when AI produces content in response to a prompt, human users do not automatically exercise the kind of creative control that copyright requires. Instead, the Office now examines, case by case, how the AI system works and how it was used, asking whether the resulting expression can genuinely be traced to an author’s “own original mental conception, to which [the author] gave visible form.” That is the line the Office has begun to mark in decisions such as Zarya of the Dawn: the comic’s written text and panel layout were protected, but its Midjourney-generated images were not.  The work here is to articulate where exactly that threshold lies and, in practical terms, to build a mechanism for transparency and clear trackability of human contribution to AI output.

4. Accept that some AI outputs will be authorless

In some cases, it may be that no one has exercised authorship in the sense we have identified. When neither user nor developer has exercised genuine creative agency, there is an argument for treating the resulting outputs as public‑domain material. Copyright was never meant to enclose every pattern or artefact: trivial lists and purely mechanical reproductions already fall outside its scope. Extending this logic to certain AI-generated outputs is an acknowledgement that authorship is something more than mere causation.

5. Avoid grouping distinct concerns under a single heading

None of the above settles the broader controversies about AI and creative practice: the rights of artists whose works are used to train models, the economic impact on creative professions, and the concentration of power in a handful of large tech companies. Conflating those debates with the question “Is AI an author?” risks distorting both. It encourages us to think of systems, rather than people, as the primary subjects of creative life, and it threatens to crowd out the claims of human authors whose works were used, often without consent, to train these models in the first place.