In January 1978, the French legislature enacted the Loi Informatique et Libertés, the statute that would govern computing and data from that point on. Its first article placed four values beyond the reach of any computing system, in this sequence: human identity, human rights, privacy, and individual and public freedoms. That sequence was a choice. Human identity came before privacy because the legislators who wrote the law understood (years before the internet, mobile computing, or anything we would now call artificial intelligence) that the deeper risk of digital technology was not to personal information but to the person.
Some almost 20 years ago, working as the employers' representative for UNIDIS in national branch negotiations for the French paper industry, I found myself at a table where that law still set the terms. We were producing a "charte informatique": informatics charter governing what computer systems could legitimately know about the workers who used them, written in the shadow of Article 1. The question of what a system could record about a person was already, in French law, a question about human identity itself.
I did not have the philosophical vocabulary for it then (practitioners rarely pause to name the thing they are doing). It found me later, in the growing gap between what the technology could do and what any existing framework (legal, ethical, or contractual) could account for: systems that do not simply process information but generate, decide, and act on behalf of the human who deployed them. The stakes embedded in a 1978 law had not shrunk. They had grown into something the law's authors could not have imagined: not what the system could do to our information, but what it could do to who we are
I brought that question to Australia.
Living here has given me a different angle of incidence on the same concern. I arrived carrying instincts formed in France: in French labour law, in the tradition of "concertation sociale", in a culture where the philosophical question and the practical question rarely stay entirely separate, where a labour tribunal will cite Sartre and a board member will speak of human dignity without feeling the need to apologise for the abstraction. Perth is far from Paris. The distance has been clarifying and it is harder to take a habit of mind for granted when the culture around you does not share it.
What I observed (and observation is not verdict) is a difference in the question being asked. Not in the sincerity or the rigour of those asking it. Australia has serious people working seriously on AI ethics and governance, and I have encountered that work with respect. The difference is prior to any answer. It concerns which question has been selected as the starting point, and what is assumed to belong to the question before anything is committed to paper.
I am French, I cannot help it. When I analyse AI in policy documents, in governance frameworks, in the board conversations I am part of through my work at Tricore Tech, I reach instinctively for the large picture before I reach for the guardrail. And this article is the result of that reflex applied to a comparison I have been carrying for several years, across two hemispheres, in two languages, from inside two professional traditions that do not ask the same questions.
That instinct has a formation behind it. The French engagement with AI's ethical stakes is a lineage, not a single voice, and it runs back further than the current debate.
Gilbert Simondon (1924-1989), the philosopher of technology whose work shaped a generation of French thinkers, argued that technical objects do not simply extend human capacities: they participate in the ongoing process by which a person becomes an individual. His theory of individuation, developed in the late 1950s and now experiencing strong academic revival in AI contexts, proposes that identity is not a fixed thing to be protected but an ongoing process to be sustained. Technical systems either support that process or interrupt it. The question Simondon asks of any technology is: what does this do to the person becoming?
Bernard Stiegler (1952-2020), another influential French philosopher, built directly on that foundation, arguing that digital systems progressively proletarianise the mind: not just labour in the economic sense, but the faculties of attention, memory, and creation that constitute human interiority. As those faculties are increasingly delegated to machines, human beings lose not only their jobs but something more intimate: the capacity to be the authors of their own knowledge and their own work. Stiegler died in 2020 and his influence on French thinking about AI has, if anything, deepened since.
Into this lineage steps Eric Sadin, whose December 2025 address at UNESCO called for what he termed a "combat civilisationnel". His most recent book, Le désert de nous-mêmes (L'Échapée, 2025), argues that generative AI represents a complete paradigm rupture: technique has reached the point where it takes charge of the intellectual and creative faculties previously distinctly human, rendering those faculties, in his own words, "presque caduques" (almost redundant). The register is different from what governance typically demands. Where risk frameworks ask what AI could do to a system or a sector, Sadin asks what it is already doing to the human person considered as a self-determining being. The distance between that question and the one most governance frameworks pose is precisely where my current thought lives.
Eric Sadin and his predecessors might read as intellectuals in conversation with themselves, were it not for what has happened inside the French state over the same period. When the mathematician and former parliamentarian Cédric Villani produced his 2018 national AI strategy report, he titled it Donner un sens à l'intelligence artificielle: not governing AI, not regulating AI, but giving meaning to it. The title declares an orientation before a single policy is proposed. France's June 2026 national plan for AI in public services is officially named "Notre IA", with an explicit designation: "une IA utile, humaine et souveraine" (useful, human, and sovereign). The word "humaine" appears not as rhetorical ornament but as an attribute the AI itself must carry, written into the official title of a government strategy document.
The same orientation has reached the school system. France's Pix IA programme has been rolling out from 2025 through 2026 as a mandatory pathway across secondary education, covering not only how AI systems work but ethics, digital sovereignty, and the development of critical thinking against manipulation. Prime Minister Lecornu's June 2026 announcement of dedicated AI teaching hours at lycée level framed the obligation precisely: "We cannot leave an entire generation to discover AI without giving them the keys to understand and therefore master it." The verb master ("maîtriser" in French) carries a specific philosophical weight in French: it is the vocabulary of autonomy. And what is being protected, in the first instance, is the capacity of the person to remain the directing intelligence in their own life.
France's critics would note the pattern: a state with a long habit of presenting its cultural assumptions as universal values and an intellectual tradition with a taste for abstraction. The observation is fair, but it does not account for where the same prior question has since travelled.
And it has travelled, among other places, to Rome. On 14 January 2025, the Vatican released Antiqua et Nova, approved by Pope Francis, warning against what it called the functionalist risk: the tendency to measure the human person by cognitive output or productive achievement rather than by inherent dignity. The same instinct that France encoded in law in 1978 was being restated, from a different tradition and a different register, by the Church. Pope Leo XIV went further. Magnifica Humanitas, signed on 15 May 2026 (the 135th anniversary of Rerum Novarum, deliberately chosen) and published ten days later, frames AI as a transformation of civilisational magnitude, comparable in consequence to the first industrial revolution. At the encyclical's presentation, Leo XIV stated that artificial intelligence needs to be "disarmed": a word chosen, in his own account, because "this moment needs words capable of attracting attention, awakening consciences and indicating paths forward." A philosophical tradition rooted in Paris and a theological tradition rooted in Rome have arrived at the same prior question by different routes. The convergence is evidence that the question is real and not just rooted in a French cultural preference.
It would be easy, and wrong, to make this into a simple contrast between depth and absence. The question has not gone unasked in Australia. Toby Walsh, Chief Scientist at UNSW.ai, has asked it most publicly. At a lecture at Newington College in August 2025, he pressed on what philosophy of mind owes to the emergence of AI: what is the nature of mind, can consciousness emerge from computation, what does it mean to be human when another kind of intelligence enters the room? His Machines Behaving Badly: The Morality of AI (Black Inc., 2022) takes these questions seriously for a broad audience. He has campaigned for autonomous weapons bans and made the case at the United Nations that AI is a moral question before it is a technical one.
Reading Walsh's practical advocacy alongside those public lectures or interviewes, a distinction becomes hard to ignore. His sustained focus is on facial recognition abuse, digital privacy violations, algorithmic assessment in schools, and robodebt: real injuries, rigorously pursued. The question he asks is what AI can do to harm us. The French tradition, from Simondon to Sadin, asks something prior: what does AI do to what we are? Walsh defends against discrete injuries to persons already formed while the French tradition contests the conditions under which persons are formed at all. These are not identical inquiries, and the gap between them determines at what point governance enters the question and what it is designed to protect.
Walsh is not alone in this and, in my opinion, the institutional framework tells the same story. The International Association of Privacy Professionals, in its November 2025 Report, observed that Australia's AI regulatory journey had shifted from an early plan to introduce an EU-style, risk-based regime toward a more flexible, standards-led approach: a recalibration driven primarily by priorities around productivity, innovation, and the use of existing legal frameworks. The Australian National AI Plan 2025, published in December 2025, opens on precisely those terms. The CSIRO's AI Ethics Framework (the framework underpinning Australia's voluntary ethics principles) anchors its philosophical grounding in existing human rights frameworks and identifies the cultural touchstone for ethical AI as the "fair go": ensuring algorithmic systems treat different groups without discrimination.
A "fair go" carries genuine moral weight (see a previous article on this concept from a French perspective). But it is an egalitarian concept, not a humanist one. It asks whether the distribution is just and does not ask whether the technique itself changes the nature of what is being distributed. A perfectly fair AI system, for example one that scores every worker by the same criteria without demographic bias, could still "proletarianise" the mind of every worker it scores, in Stiegler's terms. The fair go framework does not see that effect, because the question that would make it visible was never the starting point.
That gap is a design consequence and not a moral failure. A governance framework built without first asking what we are protecting, in the human sense, will default to harm reduction and compliance. Those are not negligible aims. Harm reduction saves people from real damage, and compliance frameworks do constrain real misconduct. But harm reduction presupposes a theory of what harms, and compliance presupposes a theory of what is owed. Without the prior question, both remain, at some level, incomplete. They govern the how without settling the why.
What the French tradition offers (and this is what I find myself carrying into Australian governance conversations) is not a set of answers but a habit of asking what a technology does to the conditions of human self-determination before asking what it does to productivity or risk, and of treating human dignity not as a constraint on AI deployment but as the criterion by which the deployment is assessed from the outset.
This does not require importing French philosophy wholesale, or establishing a national council of techno-critics, or adding a ninth voluntary principle to the eight that already exist (the French solution to everything). It requires something simpler and harder: that the people in the room (board members, governance professionals, senior executives) arrive with the prior question already formed. What kind of human activity are we protecting here? What does our organisation owe to the people inside it, not at the level of their data or their output, but at the level of their capacity to think, create, and be the authors of their own professional existence?
There is no compliance framework for these questions. They falls to the person in the room.
At a table in France, some years ago, I found myself arguing about what a computer could legitimately know about a worker. The argument felt consequential then. I could not have said precisely why.
The vocabulary to name it had been there for decades, formed long before the current debate. It took coming to Australia (with its particular focus on risk, harm, and the fair distribution of both) to make that vocabulary feel necessary rather than merely available. The gap was the occasion. I did not see it that way when I arrived. I do now.
Which is, I think, reason enough to ask the prior question today.
The vocabulary to name it had been there for decades, formed long before the current debate. It took coming to Australia (with its particular focus on risk, harm, and the fair distribution of both) to make that vocabulary feel necessary rather than merely available. The gap was the occasion. I did not see it that way when I arrived. I do now.
Which is, I think, reason enough to ask the prior question today.
The question of what AI does to what we are is real. It is the one I find myself reaching for first, with a reflex I could not suppress if I tried. If this article persuades one board member, one governance professional, one senior advisor to carry it into the room before the risk register is opened, it will have done what it set out to do.