On 23 September, Berlin-founded Noxtua announced the close of a Series C funding round of more than €100 million. The round is led by German legal publisher C.H.BECK, which has moved from Series B lead investor to new majority shareholder, with Austrian legal publisher MANZ joining as a minority investor.
For C.H.BECK, a 260-year-old family-owned publishing house, this is the largest investment in its history. Several earlier investors (Global Brain Corporation, KDDI Open Innovation Fund, CMS, Dentons and Dominik Schiener) are stepping back to make room for the publisher-led cap table, while CMS and Dentons stay on as anchor clients.
Noxtua launched its first Legal AI product in 2024, after having been founded in 2017 out of research at Oxford University and Imperial College London. Today it reports more than 30,000 users, around 100 employees, and six offices across Europe (Berlin, Paris, Stockholm, Zagreb, Munich and Fribourg). The platform covers legal research, analysis, and document drafting through jurisdiction-specific Legal AI Workspaces that run on exclusive curated content from a growing network of European legal publishers. It holds BSI C5, ISO 27001 and ISO 42001 certifications, and is engineered against the professional-secrecy and criminal-law obligations that German lawyers face under Section 203 of the Criminal Code and Section 43e of the Federal Code for Lawyers.
This week, The Legal Wire spoke with Joséphine Mansour, Chief Alliance Officer at Noxtua (and founding CEO of its French subsidiary), about what the Series C means for the company’s trajectory, why she thinks the word “sovereignty” by itself brings nothing, and the ecosystem she argues European legal AI needs if any of it is going to work.
From lawyer and GC to legal AI
Joséphine has sat on various sides of the table in her career. She practised at Freshfields and Cleary Gottlieb in Paris and New York, held senior in-house roles across industry, and spent more than five years as Senior Vice President and Group GC at SOITEC. In March 2025 she took the role of founding CEO of Noxtua France. By December, she had also taken the Chief Alliance Officer role for the group.
The move into legal AI reads, in Joséphine’s telling, as the natural end point of a career that was always drifting toward the business side. “I started in private practice as a lawyer. Then I became a Legal Director for listed companies. I wanted to move more and more to the business side.” Legal AI, she says, is where that trajectory lands now, because it changes what the job itself looks like. “It’s not another legal tech. It’s starting to change the substance of how lawyers work: how they research, how they draft, how they analyse information, make decisions and interact.”
Joining Noxtua, she is clear, was not a question of leaving the law. It was a way of staying closer to it. “Enabling lawyers and legal professionals to really focus on what matters, and helping them even more to become business partners.” The GC muscle memory is also what she thinks gives her an edge. “I know exactly what GCs are expecting from law firms, what law firms are expecting from AI, what GCs are expecting from AI. This makes me close to my customers because I feel for them. I was in their shoes.”
A short clarifying note, on why Noxtua resonated with her, in particular: she did not join just any legal AI company. “I care about trust and having a trustable and confidential answer. I don’t want to just sell a product.”
Trust that is not a slogan
Trust comes up often enough in legal AI marketing that it has started to lose its weight. Joséphine thinks it has to have meaning before the word is worth using. “Trust and responsible AI cannot simply be values that you put on a website, talk about or mention because it sounds good. They have to translate into very concrete actions.” The actions she lists are specific: how you handle the data, which models you use and how, how transparent you are with customers, who can have access to the data.
She breaks the trust question into a few layers.
First, the technology itself. Do you understand what the system can and cannot do, and are you honest about the second part? “I worry about presenting AI as something infallible. It’s not the case. It’s extraordinary, but you need, especially in Europe, to have your human judgment behind it.”
Second, the data. Noxtua does not train on or use the open web. It is grounded on curated legal data from European publishers, which Joséphine describes as “years, decades, and in some cases even centuries of research, commentaries and things that matter and explain the law.”
Third, confidentiality, security and the regulatory process, which she argues can only be made real by third-party certification rather than vendor assertion. “Everyone knows it’s not that simple” when someone tells her they have confidentiality covered but they are using standard ChatGPT. “In law, you cannot do it that way.”
One of her side projects makes the point in a different form. Since September this year, she has been teaching on the Master 2 in Artificial Intelligence and Governance at Université Paris 1 Panthéon-Sorbonne, with a focus on transparency, explainability and human oversight under the evolving French and European framework. Relating this to her role at Noxtua, Joséphine explains “these are not just words that I’m using. It’s something I feel about. It’s something that is in my DNA.”
TLW: You started teaching at Panthéon-Sorbonne this September. What are you noticing about this cohort’s instincts around AI that you wish today’s practising lawyers and GCs would adopt faster?
Joséphine: “What strikes me about this cohort is their willingness to question both AI and the way we have traditionally worked. They are preparing for a profession in which AI will be part of everyday practice. The question is how to use it intelligently while retaining independent judgment.
Teaching transparency, explainability and human oversight brings that challenge into focus: what makes an answer trustworthy? Can we verify its sources? What information are we sharing? Who remains accountable for the decision?
I would like practising lawyers and GCs to bring that same curiosity to their own workflows, alongside the experience students are still acquiring. We need to move beyond experimenting with tools and start reconsidering how legal work is organised, reviewed and valued. The essential skill is knowing how to challenge an output, recognise what is missing and decide when human expertise must take over. That deserves as much attention as learning to prompt.”
The governance behind the sovereignty label
The sovereignty framing is one Noxtua uses on its website, as do many other European legal AI companies. Joséphine provides context she says is essential to understanding the true meaning of this framing. “The thing is, being in Europe, all of this means extraterritorial laws don’t apply to us. It’s about governance.”
Her argument is that sovereignty is a label. What a customer is really buying is the structural reality underneath it. As an example, European governance means that an extraterritorial order from outside the EU cannot compel disclosure of customer data by Noxtua or its investors. European encryption and data residency mean the technical surface is also contained. “Sovereignty by itself brings nothing. It’s what’s behind it.” She also notes that even where the technology is European, the question of who owns the company matters. “If your governance is not European, your investors know so much information. If they are compelled to give it, they will give it.”
This is where the Series C changes something tangible for the company. Replacing a mixed international investor base with two independent European legal publishers moves the governance centre of gravity. C.H.BECK is family-owned and German. MANZ is family-owned and Austrian. Neither is subject to foreign disclosure regimes in the way a US or Japanese financial investor might be. For a customer worried about a 2026 version of the CLOUD Act conversation, that structural shift isn’t cosmetic.
TLW: For a GC evaluating a legal AI tool today, what are the two or three specific governance questions you would want them to ask any vendor, including Noxtua, before signing?
Joséphine: “I would ask every vendor, including Noxtua, three questions and expect evidence behind the answers.
First, where does our data go, and who can access it? That includes hosting, subprocessors, any external model providers, retention, use for training and potential access under foreign laws. Confidentiality and legal professional privilege require more than a promise that data stays in Europe.
Second, how can our lawyers verify and challenge an answer? Ask the vendor to demonstrate source traceability, how the system handles uncertainty and what happens when sources conflict. Test it on your own legal work: a convincing demonstration is only a starting point.
Third, what control do we retain throughout the relationship? That means access permissions, audit logs, notification of material changes to models or hosting, incident management and a workable exit.
The overarching question is whether the GC can explain, document and defend the organisation’s use of the tool. Certifications help, but they cannot replace that assessment or the vendor’s contractual commitments.”
The importance of the publishers
Publishers investing in legal AI is not a new story. But what’s different about the Noxtua version is the scale and the signal direction. C.H.BECK did not partner with Noxtua as a channel. It took a majority stake and MANZ followed. Joséphine is careful not to overdramatise the news, because C.H.BECK was already Noxtua’s largest investor from the Series B. “There is nothing new. And we’ve used this period since C.H.BECK first joined in the Series B to grow stronger together.” What has changed, however, is the depth of the commitment, and the market read it should prompt.
Her reading is that content is the real scarcity in European legal AI, well above model capability, and “working with publishers brings the quality.” The language of “Europe’s largest legal database” and “Europe’s largest network of independent legal publishers,” which Noxtua now uses officially, becomes more of a structural moat once the publishers themselves are on the cap table rather than merely licensed as suppliers.
The publisher roster Noxtua now draws from: Helbing Lichtenhahn in Switzerland, Blendow Group in Sweden and Ciela Norma in Bulgaria (among the content partners), alongside C.H.Beck’s affiliates in Poland, the Czech Republic and Slovakia, each bring a jurisdictional corpus that an open-web-trained model would not have in any coherent form.
The EU AI Act and the operational gap
Joséphine views the EU AI Act pragmatically. The Act, she says, has done the useful work of moving AI governance from the abstract to the operational. “Before, when people were talking about AI, they were touching on transparency, fairness, human oversight. Today, companies have to translate the EU AI Act into operational things. Who is responsible for the AI system? Companies that already have a person in charge of AI know exactly who does what. Others don’t.”
She explains that the pertinent question, in her view, is not purely based on the Act, but the maturity gap between companies. She draws the obvious comparison to GDPR, where the shape of real-world compliance only emerged once the sanctions started landing. “Let’s see when the sanctions come, how they will sanction it. With the GDPR, this is the way we were able to assess how mandatory things were, or whether best efforts was enough. We’ll know better when the sanctions come.”
She clearly states her own definition of better regulation. Not more rules, but harder ones around transparency specifically.
TLW: What is one specific thing most companies are still not doing on the operational translation of the EU AI Act, that you think the first wave of sanctions will catch them out on?
Joséphine: “I would be cautious about identifying one failing shared by most companies, or predicting what the first wave of sanctions will target. Organisations vary considerably in their uses of AI, their exposure to risk and the maturity of their governance.
Taking Article 4 as an example, the Commission’s guidance emphasises that AI literacy measures should reflect staff experience and the context in which systems are used. Penalties and other enforcement measures also depend on national law. That makes broad predictions difficult.
The practical question for a GC is therefore: have we identified the obligations applicable to our actual uses of AI, assigned responsibility and put appropriate measures into practice? That assessment needs to be specific to the organisation. I would be wary of treating either a generic compliance checklist or a prediction about enforcement priorities as a substitute.”
The thread running through Joséphine’s view of the market is that no single actor can make European legal AI work on its own. Her phrasing: “you need technology, you need publishers, you need law firms, you need corporates, you need institutions to work together.” She holds board or advisory roles at Labelia Labs and Album Paris, serves as an AI ambassador for the French Ministry, and now teaches at Panthéon-Sorbonne. The scope of what she describes as her role is wider than Noxtua.
Part of this is biographical. She spent long years on both the law firm and the in-house sides of the same table, and the two groups see the technology differently. Her practical workflow reflects that: in France she organises recurring sessions with lawyers and in-house teams to understand what, as she puts it, “gives you nightmares at night.”
Part of it is also strategic. Noxtua’s content moat is built on partnerships. Its go-to-market reach depends on bar associations, universities and innovation networks as much as on direct sales. The investor base is now publishers. The anchor clients are two global law firms. The teaching work plugs into the next generation. In a sector where many founders position themselves against incumbents, Joséphine’s version of the pitch is the opposite: a thesis in which Noxtua works because everyone else in the room is working with it.
TLW: You have described a long list of stakeholders who need to be in the room for European legal AI to succeed. Where in the European legal AI ecosystem do you see the biggest gap between appetite and action right now, and what would help close it?
Joséphine: “From my perspective, the biggest gap is between the appetite for European legal AI and the willingness to make the commitments needed to build it. There is considerable interest in tools grounded in European law, languages and professional standards. Turning that interest into products requires publishers, technology companies and legal practitioners to work together on content, workflows and validation.
That is where collaboration becomes more demanding: agreeing on licensing, allocating resources, sharing expertise and deciding who is responsible for what. Each stakeholder can see the opportunity, but their priorities and investment horizons do not always align.
What would help is more concrete cooperation around defined uses: a publisher contributing authoritative content, a technology partner building the capability, and legal teams committing time to testing and, where the value is demonstrated, adoption. Universities and professional bodies can contribute evaluation and training. European legal AI will advance through those sustained commitments, with clear responsibilities and a viable economic model.”
What AI should give lawyers back
Joséphine’s closing answer, when asked what she wants to see happen for her customers, is personal. She recalls being pregnant and billing sixteen-hour days as a GC. The point of the technology, as she describes it, is to give lawyers their time back for the parts of the job that lawyers trained for. “I want to see in-house lawyers dedicating time to being more part of the business, doing more strategy and value-added things. The same for lawyers, to be closer to their customers, thinking strategy, and not just legal. AI is not going to replace lawyers, but we need to be different.”
The Series C gives Noxtua the capital and the publisher-anchored governance to attempt that at a pan-European scale. Whether the ecosystem Joséphine describes materialises in the way she sketches it is a question enforcement of the EU AI Act, publisher consolidation and legal-market adoption will demonstrate. What is clear from the conversation is that the Chief Alliance Officer title is not an ornament. The alliances, in her account, are the product.
