Beyond the Benchmark: Who Will Control Europe's Legal Knowledge?
- Cosmonauts Team
- 2 days ago
- 7 min read
For most of the past two years, the legal profession has measured artificial intelligence by its outputs: how fast a model drafts, how rarely it hallucinates, how much time it saves. Joséphine Mansour wants the profession to ask a different question entirely. Ahead of her keynote at the Legal Innovators Europe - France in Paris Joséphine sat down with us to explain why she believes the decisive question in Legal AI is no longer which model is best, but who controls the infrastructure and thereby access through which legal knowledge flows. What follows is an edited transcript.
You're opening your keynote in Paris by telling a room full of lawyers and legal professionals to stop talking about models. That's a provocative way to start at an AI conference.
It is deliberate. When AI is discussed today, the conversation almost always starts with a one sided-view of technology, which model is most powerful or which solution delivers the highest productivity gains. Those are real and important questions. But I no longer think they are the most important ones. A year ago, law firms were still asking whether they should use AI at all. That debate is largely over; the market has moved. And once a capability becomes widely available, competitive advantage stops being about the capability itself. It shifts somewhere else, toward governance and infrastructure. Those who control the technology increasingly control access to knowledge. AI is becoming the gateway through which legal professionals access legal information.
Those who control the technology increasingly control access to knowledge. AI is becoming the gateway through which legal professionals access legal information. And the most recent developments in which technology was weaponized by turning off the access to some of the most powerful AI models, demonstrate clearly what we’ve been saying for years: European digital sovereignty must be at the core of Legal AI – or maybe even any AI used in Europe for that matter. It is dangerous if access to critical AI technologies can be restricted or conditioned by decisions taken outside of Europe.
I want the room in Paris to sit with that shift, because most procurement conversations haven't caught up to it yet.
You reach for history to make the point, energy, semiconductors. Why those analogies?
Because Europe has lived this pattern before, and we tend to recognisze it only in hindsight. Twenty years ago, energy was treated as a commodity. The discussion was about efficiency, cost, and availability. Then geopolitical events reminded us that dependency can become vulnerability very quickly. A decade ago, semiconductors were seen as technical components. Few people outside the sector had heard of ASML, a European company. Today it is among the most strategic firms on the continent, because it builds the extreme ultraviolet lithography machines without which the world's most advanced chips cannot be produced.[1] Technology became infrastructure; infrastructure became power; power became geopolitics. I believe legal knowledge is on the same trajectory. The lesson from both cases is the same: dependency often looks like efficiency, right up until the day it becomes vulnerability.
Sovereignty is the word everyone now uses. Many people assume it's a question of where the servers sit. You say it isn't.
It is not. Server location is the part of the conversation that feels tangible, so people anchor on it. But sovereignty is ultimately about legal authority, who has the power to decide, who has the power to compel, and who has the power to change the rules. Jurisdiction follows the company. Legal professionals understand this instinctively. When we structure a transaction or assess a dispute, we ask who has authority, which law applies and which court has jurisdiction. Strangely, those same questions are often absent when the profession evaluates AI. And the answer is not theoretical. In June 2025, before a French Senate inquiry on public procurement and digital sovereignty, a representative of Microsoft France was asked under oath whether data hosted on European servers could be guaranteed to stay beyond the reach of a lawful request from US authorities. The answer was no, such a guarantee could not be given.[2]
That testimony has been cited a great deal. What do you take from it?
I want to be precise about what it does and does not mean. It does not suggest misconduct, and it does not imply that governments routinely reach into confidential files. What it clarified is something more fundamental: even the most trusted company in the world cannot free itself from the laws to which it is subject. So the debate is not about whether you trust a particular vendor. It is about legal authority and governance. And in our field that distinction is important. For a lawyer, professional secrecy is only secret if you can guarantee it stays that way. The moment the guarantee depends on a foreign legal framework you cannot influence, the obligation you owe your client rests on someone else's discretion.
You frame this as a boardroom issue. Why should a General Counsel care?
Because concentration risk is a concept every executive already understands. We diversify suppliers, investments, manufacturing capacity, we treat over-dependence on a single point of failure as a discipline. Yet in Legal AI, many organisations are concentrating an unprecedented volume of legal knowledge, workflows, and reasoning inside a very small number of infrastructures, often without naming it as concentration risk at all. If a General Counsel presented that profile in any other critical part of the business, the board would ask hard questions immediately. The same discipline should apply here. This is no longer only a software decision. It is becoming a governance decision and increasingly a sovereignty decision.
Is this an argument against American technology?
No, and I am careful to say so plainly. The United States has produced some of the most remarkable technological advances of our time, and Europe benefits enormously from them. The question is whether Europe preserves its ability to make independent choices. Strategic autonomy means retaining options, maintaining resilience, and keeping the capacity to determine your own future. Those are not anti-American positions at all. They are simply the positions a serious institution takes about anything it cannot afford to lose control of.
So what does that look like in practice and how does it shape the way Noxtua is built?
We made a deliberate choice not to be one more interface sitting on top of someone else's stack. We control the system end to end, infrastructure, model architecture, data, and the application layer, and everything runs on European infrastructure operated by European partners, with no US hyperscalers. That is the structural answer to the legal-authority problem. But infrastructure alone is not enough for legal-grade work. An AI trained mostly on generic internet data cannot reach the reliability the profession requires. Additionally, the factor of legal bias comes into play: AI systems tend to reflect the data they’ve been trained on. ,If legal professionals increasingly rely on systems primarily trained on non-European legal sources, this may gradually influence legal reasoning and legal culture. and much of the legal content online reflects other jurisdictions. We wanted Noxtua to reflect the European way of legal thinking. So we build on exclusive, curated legal data from leading European publishers, integrated country by country, inside Legal AI Workspaces such as Beck-Noxtua in Germany, MANZ-Noxtua in Austri, or Swiss-Noxtua. None of that can be done in isolation, which is why alliances sit at the centre of our strategy, publishers, law firms, academic institutions, and European infrastructure partners, each contributing a piece no single actor could assemble alone.
Europe is often described as behind in this race. You sound more confident than that.
Europe possesses extraordinary assets: world-class legal publishers, world-class universities, legal professionals, and one of the most sophisticated regulatory environments anywhere. What Europe historically lacked was technological integration, the capacity to bring those assets together into a coherent system. That is changing. And I believe the next chapter of Llegal AI will not be won by the company with the largest model or the largest marketing budget. It will be won by whoever can combine technology, trust, transparency, legal expertise, and sovereign infrastructure into a single, coherent ecosystem.
You end your keynote on the word "trust." Why give it the last word?
Because in our profession trust is the product. Legal knowledge is not ordinary content. It governs contracts, litigation, compliance, public administration; it holds companies' secrets and states' agreements; in many respects it underpins the rule of law itself. The real choice in front of Europe is to prove that, in the legal sector, sovereignty can itself become a competitive advantage, that building responsibly is the more durable one. That is the argument I'm bringing to Paris.
Mansour's case lands at a moment when the abstract has become concrete. Across Europe, governments, courts, and parliaments are re-examining their reliance on foreign-controlled infrastructure, and the questions she raises, who decides, who compels, who can change the rules; are moving from conference panels into procurement files and board agendas.[3] Whether sovereignty becomes a genuine competitive advantage in Legal AI, will be decided by the institutions willing to treat it as a governance question. That is the conversation she will open on stage in Paris and, on the evidence of this exchange, it is one the profession is now ready to have.
A closing that stayed with us after this conversation was Joséphine Mansour’s insistence that the next chapter of Legal AI will not be decided solely by the sophistication of models, but by the structures that govern access to legal knowledge itself. It is a subtle shift in perspective, yet one with profound implications. In a profession built on trust, confidentiality and the rule of law, questions of infrastructure, authority and resilience are beginning to matter every bit as much as speed and productivity.
That is why we are especially looking forward to hearing Joséphine expand on these themes live in Paris. Her keynote promises to move the conversation beyond familiar debates and towards a question that may shape the future of the profession for years to come: whether Europe can turn sovereignty, trust and legal expertise into a lasting competitive advantage.
If these are questions you believe deserve a place on every legal leader’s agenda, we hope you will join us in Paris and continue the conversation with the wider Legal Innovators Europe community.
📅 June 24th - 25th, 2026
📍 Pullman Paris Tour Eiffel, Paris, France
🎟️ 𝗙𝗿𝗲𝗲 𝗳𝗼𝗿 𝗦𝗲𝗻𝗶𝗼𝗿 𝗟𝗮𝘄 𝗙𝗶𝗿𝗺 𝗮𝗻𝗱 𝗜𝗻-𝗛𝗼𝘂𝘀𝗲 𝗟𝗲𝗴𝗮𝗹 𝗣𝗿𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹𝘀:
Footnotes
[1] ASML, headquartered in the Netherlands, is the sole producer of extreme ultraviolet (EUV) lithography systems used to manufacture the most advanced semiconductors. On the narrowing of performance differences between leading AI models, see Stanford Institute for Human-Centered AI (HAI), AI Index Report 2025.
[2] Testimony of a Microsoft France representative before the French Senate inquiry on public procurement and digital sovereignty, June 10, 2025 (transcript available at senat.fr). Reported in "Not Sovereign: Microsoft Cannot Guarantee the Security of EU Data," heise online, July 2025.
[3] For an overview of the wider European migration away from US-controlled infrastructure — including decisions in Denmark, Switzerland, France, and at the International Criminal Court — see Paula Reichenberg, "The Great Divorce," Noxtua blog (noxtua.com/company/blog/the-great-divorce).
Comments