The case of Thélyson Orélien now exceeds the mere question of whether a writer has used artificial intelligence or not. It reveals a deeper difficulty: we possess tools capable of directly participating in writing, yet we do not yet have sufficiently precise categories to distinguish assistance, mediation, delegation, and substitution.
The controversy surrounding Thélyson Orélien's novel It Was That or Die has shifted dimensions. After accusations of using artificial intelligence, the author's and publisher's denials, and the emergence of plagiarism allegations, the Académie Goncourt decided to withdraw the book from its selection. It asserts in particular that it wants to preserve the integrity of the prize and not endorse the creation of texts generated "with the help of AI". This institutional decision is significant, but it does not resolve the conceptual problem. What exactly does it mean to write "with the help of AI"? Requesting a translation, organizing notes, comparing several formulations, testing an argument, producing a paragraph, or entrusting a system with the substantial writing of a work are not equivalent gestures. Yet our public language often continues to group them under a single formula: "using AI".
I am not seeking here to exonerate Thélyson Orélien nor to pronounce his guilt. The making of his novel is a factual question. What interests me lies rather at the margin of the alternative between accusation and defense: what this case reveals about our difficulty in thinking through new mediations of writing.
Thinking has never been an act without mediation
We think with a language we have not invented. We inherit concepts, narratives, categories, and problems; we read, take courses, discuss with professors, encounter objections, and exchange with others. An idea we consider personal may have been prepared by an old reading, a forgotten conversation, or an experience that had not yet found its words. Personal thought is thus not a thought without external provenance.
The history of intellectual techniques also shows this. In Phaedrus, Plato already staged the ambivalence of writing: what allows us to preserve memory also transforms the way that memory is exercised. Artificial intelligence is obviously not writing, but the question remains troubling: what becomes of a human capacity when a technique allows part of it to be externalized? Autonomy cannot simply be defined as the absence of mediation. The decisive question becomes rather that of appropriation: under what conditions can we say that a received, discussed, or technically mediated thought is nonetheless our own?
Receiving an idea does not yet mean appropriating it. A student can perfectly repeat their professor's words without understanding what is being said, while an idea from elsewhere can become deeply personal when questioned, confronted with other experiences, displaced, and transformed. The question is thus not only about where an idea comes from, but what we have made of it.
This is where judgment appears to me to be crucial. In the Fourth Meditation, Descartes reminds us that we can assert more than what we actually understand. This difficulty takes on a new dimension with AI: a text can be elegant, coherent, and argued while largely exceeding the understanding of the one who recites it. The problem does not simply begin when a proposition comes from an external tool; it begins when we make our own what we are no longer able to judge.
Kant allows us to extend this reflection. His Sapere aude, the courage to use one's own understanding, does not mean that one should produce all the content of their thought alone. Thinking for oneself does not consist in living without texts, masters, traditions, or interlocutors, but in not abandoning the exercise of judgment. Spinoza helps to avoid the image of an absolutely sovereign subject: we are always affected and determined by relationships that we do not completely control. Intellectual autonomy could then be defined not as the absence of mediation, but as the capacity to understand, evaluate, transform, and take responsibility for what we make our own.
AI as incapacity: a presupposition to be questioned
One of Thélyson Orélien's first responses is, from this perspective, particularly revealing. He stated that he did not see why he would call upon a machine since he had always loved to create and write. His publisher offered a similar reasoning: he "did not need AI". These statements should initially be heard as personal denials. Nevertheless, they reveal a more general representation: using AI would primarily indicate that something is lacking.
But why should one have to "need" artificial intelligence to use it? A driver can perfectly know their route and consult a GPS; a person capable of performing a calculation can use a calculator. The tool does not solely serve to address an incapacity. It can speed up an operation, allow a comparison, open a possibility, or modify the conditions under which a skill is exercised. Thus, saying "I do not need it" does not allow one to conclude: "I would have no reason to use it".
An asymmetry then appears. Artificial intelligence is presented in many areas as a matter of scientific, economic, and strategic power. States, businesses, and institutions invest to develop their capabilities and avoid relying on technologies produced elsewhere. When AI touches on productivity, research, or international competition, it appears as a resource that should be mastered. However, when it enters certain literary or intellectual spaces, the vocabulary changes: what elsewhere represented a new capacity can become here a sign of ease, incapacity, or fraud.
It is obviously not a matter of demanding the same rules across all domains. A literary work, a university examination, and an industrial application do not engage the same responsibilities. But this difference does not prevent questioning the distribution of the legitimacy of the tool. We are in the process of integrating AI among the important technologies of our societies while demanding, in certain spaces, that the user can almost demonstrate that they could have done without it.
This situation produces another contradiction: we demand transparency while making the admission of use an immediate source of suspicion. Under these conditions, the pertinent question may not be whether an author "needed" AI, but what they entrusted to it and what they continued to take responsibility for themselves.
Differentiating use, delegation, and substitution
However, it would be too simple to conclude that artificial intelligence changes nothing. It crosses an important threshold: generative systems can produce language, propose a structure, compare arguments, develop a synthesis, and offer a response that already possesses the external form of a thought. Mediation thus reaches operations that we directly associate with the author.
This is why we need more precise distinctions. There is consultation, assistance, reformulation, partial generation, delegation, and, at the extreme, substitution. The difference between asking a system to confront a hypothesis and having it produce almost an entire text that one then signs is considerable. The phrase "he used AI" teaches us very little as long as we do not know what was entrusted to the tool.
A subject can consult a system, contest its responses, identify its errors, verify its propositions, reject certain directions, and transform what is suggested to it. Conversely, someone can receive a complete argument and accept it because it seems convincing without being capable of reconstructing its logic. In this latter case, the problem is no longer just that the machine has produced sentences: the subject has abandoned a part of the judgment they nevertheless claim to own the result of. The question then becomes less about whether AI intervened than about determining what part of the judgment has been delegated to it.
This distinction does not entail making artificial intelligence an innocent tool or celebrating its uses indiscriminately. It can create forms of dependence, homogenize writing, produce an illusion of mastery, and gradually lead to delegating operations that were previously part of the very work of thought and creation. The difficulty is precisely to identify the moment when assistance becomes dispossession of judgment. Recognizing these risks does not justify placing all uses under the same suspicion; instead, it makes the distinction between what the tool facilitates, what it transforms, and what it eventually replaces all the more necessary.
This is also where the technical question becomes political. Tools never enter a neutral space: institutions decide what is permitted, valued, suspected, or prohibited. The same use can be presented as innovation in a company and as a suspicion of fraud in a university or literary space. This distribution also concerns credibility. Not everyone has the same resources when an accusation arises. An author published by a major publishing house and recognized with awards possesses more symbolic capital than a student, a novice researcher, or a writer without an institution behind them. Bourdieu allows us to precisely understand that recognition is never solely individual: it is built through mechanisms of consecration that can also be withdrawn.
The Orélien case illustrates this particularly clearly. An institution can contribute to consecrating an author, then withdraw its recognition when it believes its own criteria of legitimacy are threatened. The question then concerns not only the writer but also those who have the power to recognize writing as legitimate: who can use the tool, who can acknowledge having used it, who will be believed, and who will benefit from the doubt?
From suspicion to responsibility
The "author function" questioned by Michel Foucault takes on a particular relevance here. The author is not just the one who materially produces each sentence; they are also the one to whom a text is attributed, the one who signs it, and the one who must be accountable for it. AI thus requires us to more clearly distinguish the material production of language from intellectual responsibility.
This obviously does not mean that anyone can sign any text. A criterion must be maintained. I would suggest searching for it in the responsibility of judgment: to be an author would mean remaining capable of understanding what one asserts, explaining why one retains it, recognizing what must be rejected, verifying important elements, and publicly assuming the result. Such a definition does not diminish the author's responsibility; it increases it.
How then to prevent every suspicion of AI from turning into a scandal? Certainly not by simply demanding "pure" texts. Rather, we should build graduated rules: distinguish technical assistance, research, reformulation, partial generation, and substantial delegation; specify what is authorized, what must be declared, and what becomes incompatible with the conditions for attributing a work or evaluating a task. Detectors should never be the sole proof. When serious doubt arises, the production process — drafts, versions, corrections, editorial exchanges, author explanations — is much more instructive than a percentage produced by software.
Above all, transparency will only function if it ceases to be automatically assimilated to a confession of guilt. A norm that punishes admission necessarily produces opacity. We should therefore move from a morality of suspicion to an ethics of responsibility and traceability: no longer simply asking "did you use AI?", but for what purpose, to what extent, what have you verified, what have you transformed, and of what do you ultimately accept responsibility?
The Thélyson Orélien case may yet see further developments, but it has already demonstrated that the opposition "human or AI" is too poor to describe the new conditions of writing. We must distinguish mediation and substitution, assistance and dispossession, use of a tool and abandonment of judgment. Intellectual autonomy has never meant absence of mediation; it perhaps consists in the capacity to remain the subject of one's judgment through the mediations that make our thought possible.
Artificial intelligence does not entirely create the problem of the author. It makes visible an old question and pushes it to a new limit: how does something that partly comes to us from elsewhere become sufficiently our own so that we can take responsibility for it? The decisive question may not only be whether a machine can write, but what operations we must continue to assume to be able to legitimately say that a word is our own. And behind this inquiry lies another, even more political: who will decide how these new capabilities will be distributed, authorized, and recognized — and who will be granted the right to use them without losing their legitimacy?
Bibliographic references
Bourdieu, Pierre, The Rules of Art. Genesis and Structure of the Literary Field, Paris, Seuil, 1992.
Descartes, René, Meditations on First Philosophy, followed by Objections and Replies, Paris, J. Vrin, coll. "Bibliothèque des textes philosophiques", 1970.
Foucault, Michel, "What Is an Author?", in 1969: Michel Foucault and the Question of the Author, text presented and commented on by Dinah Ribard, Paris, Honoré Champion, coll. "Textes critiques français", 2019.
Kant, Emmanuel, What Is Enlightenment? translation by Jean-François Poirier and Françoise Proust, dossier by Matthieu Haumesser, Paris, Flammarion, coll. "GF Philo", 2020.
Plato, Complete Works, edited by Luc Brisson, Paris, Gallimard, coll. "Bibliothèque de la Pléiade", 2008.
Spinoza, Baruch, Ethics, translation, introduction, and index by Robert Misrahi, Paris, Éditions de l’Éclat, 2005.
Press sources
Arnould, Frédéric, "The Thélyson Orélien Case, a Seismic Event for the Publishing World?", Radio-Canada, September 2026.
A.V. with AFP, "Thélyson Orélien Case: The Goncourt Withdraws the Novel from Its Selection", 20 Minutes, September 25, 2026.
Etancelin, Valentin, "Thélyson Orélien Denies Any Use of AI and Denounces 'Numerous Racist Attacks'", Le HuffPost, September 22, 2026, updated September 23, 2026.
