Deemed Authorship in the Age of AI: How Ghost Systems Undermine the Legal Myth of Creativity: A Research Dossier Alexandra Kitty

Copyright law’s foundational concepts, “author,” “original work,” and “modicum of creativity”, were constructed for a pre-digital, pre-AI world and have been quietly stretched to accommodate ghostwriting, industrial publishing, and corporate authorship through a system of legal fictions. The arrival of generative AI does not create a new crisis; it exposes an old one. The same industry that normalizes ghost teams, moral-rights waivers, and brand-managed prose now invokes a pristine notion of human creative genius to exclude AI from copyright protection. This essay argues that the contradiction is untenable, that it actively harms genuine solitary authors who pay a real market price for their independence, and that an AI defendant who pressed these inconsistencies to their logical conclusion would force courts to choose between two equally uncomfortable outcomes: radically tightening originality standards across all human works, or honestly acknowledging that authorship has always been a pragmatic legal fiction: one that AI-assisted workflows merely make visible.

I. The Wrong Operating System for a New World

Copyright’s core vocabulary: “author,” “original,” “independent creation”, crystallized in an era of print culture and individual craftspeople. The Berne Convention of 1886, the U.S. Copyright Act, and Canada’s Copyright Act all presuppose a recognizable human author whose creative choices can, at least in theory, be traced back to a single originating mind. The system was never designed to adjudicate works assembled by ghost teams, plotters, heavy-handed editors, and packagers operating under non-disclosure agreements, and it was certainly never designed to reckon with large language models trained on hundreds of billions of human-produced tokens.

Instead of acknowledging this mismatch and rewriting doctrine from the ground up, courts and legislators have applied patches: work-for-hire rules, rebuttable presumptions of authorship, joint-authorship provisions, and the “modicum of creativity” standard from Feist Publications, Inc. v. Rural Telephone Service Co.. Each patch resolves the immediate case while deepening the underlying contradiction. The result is a legal operating system running software it was never designed to handle, crashing most visibly now that AI has entered the environment.

II. The Legal Fiction of Deemed Authorship

What Statutes Actually Say

Berne and national copyright statutes handle the authorship question through a blunt presumption: the person whose name appears on copies of a work is presumed to be its author and the holder of its rights, unless someone demonstrates otherwise. This presumption is explicitly rebuttable, meaning it is not a factual claim but a procedural convenience. Scholars have called the results “deemed authorship”, a deliberate legal fiction that separates the entity entitled to enforce copyright from the human or humans who actually performed the creative labor.

Work-for-hire rules compound the fiction. Under these provisions, an employer or commissioning party may be treated as the statutory “author” of a work even though every word was written by a hired hand. The ghostwriter produces the text; the contract makes the brand the author. The law does not regard this as deceptive. It treats it as a feature: a mechanism for allocating economic rights in complex creative industries without requiring courts to audit who actually held the pen.

The “Modicum of Creativity” Standard

Feist established that copyright requires “independent creation plus a modicum of creativity”: a standard explicitly characterized as “extremely low” and resistant to precise definition. “Independent creation” does not mean “created without influence by others or by prior works.” It means only that the work was not directly copied from a specific prior text. Two authors can produce identical poems independently, and both will be “original” under this standard. A work saturated with genre tropes, shaped by heavy editing, and structured according to mandatory style guides still qualifies so long as some human(s) assembled it rather than photocopying it from an earlier work.

Courts try to keep this standard meaningful through two filtering doctrines: scènes à faire (stock elements required by a genre are not protectable) and merger (when an idea can only be expressed in a limited number of ways, the expression merges with the idea and loses protection). In practice, these doctrines strip away the enormous bulk of genre conventions, house styles, and formulaic structures before asking whether anything distinctively expressive remains. What remains after filtering is often very thin: a specific configuration of otherwise unprotectable elements and protection extends only to near-identical copying of that configuration, not to the patterns and conventions themselves.

III. How Publishing Manufactures “Authors”

The Ghost System

The commercial publishing industry has long operated on a division of labor that bears little resemblance to the romantic authorship narrative invoked in copyright litigation. Big-name authors routinely work through teams of plotters, ghostwriters, developmental editors, and production partners; non-disclosure agreements and moral-rights waivers are standard instruments for keeping actual contributors invisible. The marquee name, the brand, is marketed to readers as the singular creative intelligence behind a series. Behind the brand is something closer to a film production unit: specialization, delegation, and managed output calibrated to reader expectations.

Moral rights, where they exist in law (as in Canada and most of Europe), are routinely waived in commercial publishing contracts, permitting publishers to abridge, revise, and repackage works while maintaining the brand identity intact. An author who waives moral rights surrenders the right to object to alterations and, in many cases, the right to attribution in subsequent editions. The “author” whose name appears on the cover may have no ongoing relationship to the text at all.

What Fans Actually Buy

A useful empirical test of this system is audience perception. Superfans, the readers most invested in a particular author’s “voice”, generally cannot detect changes in ghost personnel when a series transitions between writers. Their loyalty is to a constructed brand voice, not to a verifiable individual creative signature. If those same fans were asked directly whether they would have purchased the books knowing they were produced by a rotating team rather than the named author, the likely answer, no, signals that the marketing representation, not just the legal label, is doing significant work.

This dynamic is visible at scale in film and television pitching and in book market days at major festivals. Pitches cluster around a small number of successful templates; “the next [established author]” is offered as a selling point rather than an admission of derivativeness. The genre space is effectively a parameter range: a small set of character types, structural beats, and emotional arcs within which individual projects are differentiated by surface variation rather than foundational invention.

Style Guides, Institutional Constraints, and the Loss of Voice

Outside commercial fiction, nonfiction authorship faces its own set of constraints that narrow the claim to individual creative expression. Academic writing is governed by mandatory citation and formatting systems, APA, Chicago, MLA, and structured by funder requirements, journal templates, and peer-review conventions that standardize not only format but argumentation style and acceptable rhetorical moves. Journalism operates under house styles enforced by editors and, in much wire-service and beat reporting, depends heavily on press releases and publicist-supplied materials that the journalist is expected to verify, contextualize, and reframe rather than originate. Business and legal communications are explicitly designed for interchangeability: the HR rejection letter, the boilerplate disclaimer, the standard contractual clause.

Copyright doctrine nominally accounts for these constraints through idea-expression, merger, and scènes à faire. In practice it does so inconsistently, often protecting works whose claim to original expression, honestly examined, amounts to little more than compliance with an institutional template. The “author” in these contexts is not originating a voice but channeling a format.

IV. AI as Mirror, Not Monster

What Generative Models Actually Do

Large language models are trained on vast corpora of human-produced text, learning statistical patterns: tropes, structures, phrasings, genre conventions, argumentative forms, and recombining them in response to prompts. This is, at a formal level, not categorically different from what a ghost team does when it studies an established author’s catalog, internalizes her characteristic moves, and produces new work in her style. Both processes involve pattern extraction from prior text, constraint satisfaction (genre expectations, editor preferences, market positioning), and iterative refinement by a supervising human.

Recent research makes the continuity of training data even more explicit. Ko et al. (Virginia Tech and Cisco Research, arXiv:2511.05518) demonstrate that large language models emit verbatim training data when placed in states of sustained token-level uncertainty through what they term Confusion-Inducing Attacks. On LLaMA-2 70B, 22.2 percent of attacks yield verbatim training excerpts; on alignment-tuned LLaMA-3-Instruct 70B, approximately 6 percent verbatim and 19 percent near-verbatim output still occurs. This demonstrates that models do not merely “learn statistics” but retain specific expressive sequences, a finding with real evidentiary significance for copyright litigation.

At the same time, the methodological point cuts both ways. If what is extracted under adversarial conditions is formulaic genre prose, boilerplate business language, or press-release copy, the copyright significance of that extraction is limited precisely because those genres are already minimally creative under existing doctrine. The strongest copyright claims attach to longer, distinctive, less-constrained expressive passages, and it is there that the AI copyright debate has genuine traction.

The Double Standard in Current Litigation

Courts and the Copyright Office have, in recent years, moved to affirm a human-authorship requirement as a bright line against AI copyright claims. Appellate courts have reaffirmed that works generated without sufficient human creative control are not protectable, regardless of how sophisticated or impressive they appear. These holdings are presented as protecting the human creative spark from mechanical displacement.

The same courts have not addressed, because no litigant has yet forced them to, the question of whether the “human authorship” invoked to exclude AI is consistent with the ghosted, packaged, morally-waived human authorship that copyright already protects. If the law is comfortable treating a brand-managed, team-produced, NDA-shrouded series as the legitimate copyright of its named author, it cannot coherently invoke a pristine ideal of individual human creativity to draw a categorical line against AI-assisted workflows under meaningful human direction. The contradiction has not yet been pressed in a single, integrated argument. That is the window.

V. The Unified Argument: A Litigation Narrative

Step One: Deconstruct the Plaintiff’s Authorship

Using discovery, an AI defendant targets the marquee names in the plaintiff class: the authors with the largest catalogs and the most aggressive copyright claims. Through deposition and document production, the defense elicits: the names and roles of all ghostwriters, plotters, developmental editors, and production partners who contributed to works in question; the NDA and moral-rights-waiver provisions governing those contributions; and the contractual mechanism by which the brand name became the legal “author” over those contributions. The defense then presents fan-study evidence showing that readers cannot distinguish books from different production periods or different ghost personnel, demonstrating that the “authorial voice” they believe they purchased is a brand construction, not a traceable individual creative signature.

At the same time, the defense introduces a corpus of submitted scripts, pitches, and genre works demonstrating industrial sameness: books that are parameter variations on a dominant predecessor author (a market-day full of Alice Munro permutations; a development slate of structurally identical screenplays). This evidence supports the scènes à faire and merger arguments, the “creative residue” in these works, once generic structure and tropes are stripped, is extremely thin, but it also supports the broader narrative that “authorship” in these markets is already brand management over a small creative parameter space.

Step Two: Connect to Doctrinal Fictions

Having established what the plaintiffs’ authorship actually looks like in practice, the defense connects this to what doctrine already accepts: the named “author” holds copyright not because she factually generated every creative choice, but because presumptions, assignments, and work-for-hire rules allocated ownership to her as a legal matter. The originality requirement does not probe collaborative inputs or derivative structures; it asks only whether the work as a whole was not mechanically copied from a specific prior text.

The defense then argues: this is a system that has always separated factual creative contribution from legal authorship, always accommodated derivativeness and collaboration within a low originality threshold, and always used contractual allocation rather than creative forensics to determine who can sue. The system knows this, has chosen this, and has built commercial industries on this. It cannot invoke a rigorous standard of authentic human creativity in AI cases without first applying that same standard to the ghost system it already accepts.

Step Three: Confront the Carve-Out as Structurally Incoherent

The defense now presents the court with the logical choice it has been avoiding. If the existing ghost system satisfies “human authorship”, despite distributed, anonymous, contractually suppressed creativity; despite moral-rights waivers that allow unlimited alteration; despite fan evidence that no coherent individual voice exists, then the question of what makes AI-assisted production different is not answered by invoking “the human spark.” The spark is already diffuse, contractually rerouted, and empirically indistinguishable from assembly-line output.

To exclude AI while including the ghost system, a court must articulate a principle that distinguishes them. The candidates are:

AI produces output without any human creative direction. This is empirically false for most commercial AI workflows, which involve substantial human prompting, selection, editing, and judgment.
AI training on copyrighted works is itself infringement, regardless of output. This is a separate and independently contested question; it does not bear on whether outputs are protectable.
Human authorship is a policy choice to incentivize human creators. This is the most honest framing, but it concedes that “the human spark” is not a metaphysical fact but a policy preference, one that courts would need to explicitly weigh against the costs of validating an industry that has commercially hollowed out the very authorship it now defends.

Step Four: Force the Choice

The court must ultimately choose one of two paths:

Path A: Tighten originality across the board. Require genuine, traceable individual human creative contribution as a condition of copyright. Apply this standard to ghost-authored books, heavily edited academic work, press-release journalism, and industrial scripts. The result would be a significant narrowing of protectable works across the entire creative economy, outcomes the publishing and entertainment industries are utterly unprepared to accept.

Path B: Acknowledge authorship as a pragmatic fiction. Confirm that copyright uses authorship and originality as low-bar, contractually flexible, policy-driven constructs, not as factual records of creative genesis. Then reason, consistently, that AI-assisted production under sufficient human direction and oversight falls within the same pragmatic framework, not because AI is human, but because the system has never actually required the “author” to be the full creative origin of the work.

Neither path is comfortable. Path A disrupts the existing industry; Path B disrupts the exclusion of AI. But the argument forces courts to confront the contradiction rather than leaving it unexamined.

VI. Coda: The Price of Genuine Authorship

There is a personal dimension to this argument that gives it its moral force. The analysis above is not written to defend corporate AI deployment or to trivialize the concerns of working writers. It is written from the position of an author who writes alone, no ghostwriters, no beta readers, no editors who touch the text, and who has paid a real market price for that independence. In a publishing economy organized around brand management, genre compliance, and team production, the solitary eccentric voice is commercially disadvantaged. The very system that now claims to speak for “human creativity” in AI litigation is the system that has historically marginalized exactly the kind of unmediated individual authorship it invokes.

If copyright’s romantic ideal of the solitary author has any normative force at all, it should be used to reform the practices that have already gutted it from within: not to draw a line that protects brand-managed ghost teams while excluding AI, as though the former represents authentic creativity and the latter represents its antithesis. The honest argument for protecting human authorship from AI would require simultaneously protecting human authorship from the ghost economy. No party in current AI litigation is making that argument, because no party in current AI litigation wants to open that particular door.

What AI does, ultimately, is hold up a mirror to a system that has been comfortable with its own fictions for a very long time. The question courts must now answer is not whether AI crossed some inviolable line of human creativity. It is whether that line was ever drawn where we said it was, or whether it was always just a convenient marker that moves to protect whoever can afford the litigation.

Conclusion

The legal concept of “the author” was always a construct, useful, adaptable, and deliberately detached from the messy reality of how creative work is actually produced. Ghost systems, moral-rights waivers, work-for-hire contracts, and brand management have operated within this construct for decades, with full legal sanction. The arrival of generative AI does not rupture this system; it makes its contradictions impossible to ignore.

An AI defendant who runs the full, integrated argument, from the legal fiction of deemed authorship, through the industrial reality of ghost-produced creative work, to the incoherence of invoking authentic human creativity to exclude AI while blessing the ghost economy, puts courts in a position from which there is no comfortable exit. Courts can retreat into unprincipled line-drawing, pretending the contradiction does not exist. Or they can confront it, and in doing so, either honestly narrow copyright protection for industrial human works or honestly acknowledge that AI-assisted production under human direction fits within the same legal fiction that has always governed authorship.

Either outcome is more honest than the current trajectory. And the argument is available to any lawyer willing to build it as a single narrative rather than a series of isolated doctrinal skirmishes.

Additional research (and lots of lippy back and forth) from Perplexity.