Synthetic voice · AI-generated (text-to-speech).
On 23 September 2026 it emerged that Berlin-based legal AI provider Noxtua SE had closed a funding round of more than 100 million euros. What is unusual for the German-speaking legal market, however, is not the amount but who will hold the majority: C.H.Beck, Germany’s largest legal publisher — together with the Austrian legal publisher MANZ, which joins as a new investor.
The interesting part, then, is the structure. Under the name Beck-Noxtua, Noxtua operates one of the two most widely used legal AI tools in the German market. Such a tool quotes commentary literature, the annotated treatises German lawyers cite by default — and the publisher of that literature is now its majority owner. For law firms this is not industry news but a question of dependency: if source and tool come from the same house, who is left to check what the tool cites?
At a glance
- What happened: a Series C round of more than EUR 100m; C.H.Beck becomes majority shareholder, MANZ joins, and CMS and Dentons exit as shareholders while remaining anchor customers.
- What follows: six days earlier, the Legal Tech Monitor 2026 reported usage figures for the German-speaking region — Legora 35 %, Beck-Noxtua 34 %, Harvey 29 %, Libra 27 %, “Claude for Legal” 14 %.
- The catch: the monitor’s sponsors include C.H.Beck. That does not make the figures wrong, but it does make them non-independent — and they are mentions with multiple selection, not market shares.
What emerged on 23 September
Noxtua is no newcomer: a 2017 spin-off out of Oxford and Imperial College, with, by its own account, more than 30,000 users. It runs on European infrastructure without US hyperscalers. The valuation was not disclosed; what shifted is who owns the company:
| New or strengthened | Exiting as shareholders |
|---|---|
| C.H.Beck — from lead investor in the Series B (spring 2025, EUR 80.7m) to majority shareholder | CMS and Dentons — but they remain anchor customers |
| MANZ (Austria) — newly on board | Global Brain, KDDI Open Innovation Fund, one business angel |
The transaction does not stand alone. A day earlier, Bloomberg reported that Swedish vendor Legora was in talks for at least 300 million US dollars at a valuation of roughly 8.5 billion. In August, JUVE had named the direction of travel: legal AI is becoming “the operating system of the law firm” — standalone tools displaced by integrated products that bring together matter context, internal know-how and legal reference sources. That is exactly the point: the checking function then no longer sits outside.
The figures that came just before — and who collects them
On 17 September 2026 the Legal Tech Verband Deutschland (German Legal Tech Association) presented the Legal Tech Monitor 2026: around 360 respondents, plus roughly 30 supplementary interviews. It will be quoted everywhere over the coming months, so the detail matters.
| Most-used legal AI tools | Mentions |
|---|---|
| Legora | 35 % |
| Beck-Noxtua | 34 % |
| Harvey | 29 % |
| Libra | 27 % |
| “Claude for Legal” | 14 % |
Three qualifications belong with those numbers, and none of them makes the numbers wrong.
First: the publisher is a market participant. C.H.Beck is among the sponsors. Beck’s product sits in second place at 34 percent, and the fullest public account of the results appears on beck-aktuell, the trade portal of the same house. This is not an accusation: association studies are almost always sponsor-funded. It means the choice of questions and the presentation carry the same reservation as any vendor-adjacent study. Anyone quoting the figure should quote the constellation with it.
Second: these are not market shares. The five values add up to 139 percent, so multiple selection was allowed — a firm with three licences appears three times. That is usage spread, not market share, and not revenue share. There is also an open question: only 61 percent of respondents hold a licence at all. Whether the 34 percent refers to all 360 or so respondents or only to that subset is not stated in the reporting.
Third: around 360 respondents are an indication, not representativeness. For scale: as of 1 January 2026 there were 138,420 admitted lawyers in Germany. How the sample splits between law firms and in-house legal departments is not disclosed — even though the questions address both sides: only the client side can say that 57 percent are moving work in-house, only the firm side that 39 percent have adjusted their billing model.
Consolidation seen from inside: several forms have become one. What disappears is not capacity but the space between them — the place where, until now, someone checked from the outside.
The commercial pressure, read soberly
The same monitor supplies the commercial reason why smaller practices cannot wait this out — two figures of unequal weight. 73 percent would consolidate mandates with firms that show proven AI cost savings — the figure that ends up in slide decks, and the weaker of the two: an intention under a condition, not behaviour.
40 percent require evidence on legal tech and AI when selecting a firm — an established requirement: failing it keeps you off the invitation list. Together with the 57 percent above: no reason to panic, but a reason to settle the vendor question now rather than under time pressure.
What this means for your firm
First: ask for paper, not for trust. For matter-related AI use, § 43e BRAO (Bundesrechtsanwaltsordnung, the Federal Lawyers’ Act) requires a documented selection decision and a contract in written form that binds the service provider to confidentiality and references § 203 StGB (the criminal provision on breach of professional secrecy). § 203 StGB calls for its own, criminally enforceable confidentiality undertaking — in addition to the Auftragsverarbeitungsvertrag (data processing agreement under the GDPR), which governs something else. Anyone who presents a data processing agreement and considers § 203 handled has not read the provision. The clauses in detail are in AI contracts: what really needs to be in them; the professional-rules side is in AI for law firms.
Second: ask to see the chain and the location. The vendor is not the question — its subcontractors are. The core practical problem is the US CLOUD Act as soon as a US hyperscaler sits anywhere in the chain. Here Noxtua has a genuine argument: European infrastructure without US hyperscalers. It becomes a reliable one only once it is contractually warranted, covers the entire subprocessor chain, and changes are notifiable. Ask for the list, not the statement.
Third: negotiate the exit while you still have negotiating power. Integrated products bind harder than standalone tools because they absorb know-how that then no longer sits with you alone. Settle in advance the format and deadline for data return — and whether learned firm know-how feeds model training, and can be inspected and deleted. Harvey committed to exactly that for its Memory feature in August 2026; that is the benchmark for others. The mechanics are in Avoiding AI vendor lock-in.
And why is “our publisher is trustworthy, surely” not an answer? Because reputation is not a contractual warranty: § 203 StGB requires an undertaking, § 43e BRAO a documented selection decision — documented means auditable, not plausible. And because the second reason sits in the announcement itself: CMS and Dentons were shareholders in this vendor in spring 2025; in September 2026 they are not. Relying on an ownership structure means relying on a snapshot.
Conclusion
The consolidation has its good sides. A tool that incorporates licensed commentary literature is preferable to a general-purpose AI that invents citations freely — and European infrastructure without US hyperscalers solves a real problem under the professional rules, not an imagined one.
The price is the vanished second opinion. According to the database maintained by Damien Charlotin, by early September 2026 there were roughly 2,000 court decisions worldwide with established fabricated citations, eleven of them from Germany. The most-cited German reference case — Amtsgericht Köln (Cologne Local Court), order of 2 July 2025, case no. 312 F 130/25 — involved a pleading in which every citation from page 8 onwards was fabricated, including a BGB commentary quote attributed to the wrong author. The duty to check therefore stays where it always was — backed by § 43a(3) BRAO and by liability. It does not get lighter when the source and the tool share an owner. Only less conspicuous.
If you have a vendor contract in front of you and want to know what § 203 StGB concretely requires in it, let’s talk. I read such contracts as a business lawyer and build software myself.
FAQ
What emerged about Noxtua on 23 September 2026?
Berlin-based Noxtua SE closed a Series C round of more than 100 million euros. C.H.Beck, already lead investor in the Series B in spring 2025 with 80.7 million euros, becomes majority shareholder; Austrian legal publisher MANZ joins as a new investor. The law firms CMS and Dentons exit as shareholders and remain anchor customers. The valuation was not disclosed.
What are the market shares of law firm AI vendors in Germany?
There is exactly one current source: the Legal Tech Monitor 2026 from the Legal Tech Verband Deutschland (German Legal Tech Association), presented on 17 September 2026 with around 360 respondents. It reports Legora at 35 percent, Beck-Noxtua at 34, Harvey at 29, Libra at 27 and “Claude for Legal” at 14 percent. Those are mentions with multiple selection allowed, not market shares — the figures add up to 139 percent. And the monitor’s sponsors include C.H.Beck, whose product sits in second place.
Which questions should a law firm put to an AI vendor?
Four, all in writing. First, the criminally enforceable confidentiality undertaking required by § 203 StGB — not the Auftragsverarbeitungsvertrag (data processing agreement under the GDPR), which governs something else. Second, the full subcontractor chain that § 43e BRAO wants documented. Third, data location, governing law, and whether the US CLOUD Act reaches any party involved. Fourth, the format and deadline for getting your data back on exit.
Do small law firms now have to deploy AI to keep their mandates?
The pressure is documented, but weaker than it sounds. Per the Legal Tech Monitor 2026, 73 percent would consolidate mandates with firms that show proven AI cost savings — an intention under a condition, not observed behaviour. The heavier figure is that 40 percent already require such evidence when selecting a firm. Doing nothing means losing pitches quietly rather than losing mandates suddenly.
Sources — as of 23/09/2026
- Noxtua Series C, change of shareholders (23/09/2026) — https://www.trendingtopics.eu/noxtua-legal-ai-startup-geht-mehrheitlich-an-rechtsverlage/
- Legal Tech Monitor 2026, key figures (17/09/2026) — https://www.beck-aktuell.de/rechtsbranche/anwaltschaft/legal-tech-monitor-2026-ki-tools-2026-09-17
- Legora funding talks (22/09/2026) — https://www.bloomberg.com/news/articles/2026-09-22/legora-in-talks-to-raise-funds-at-a-roughly-8-5-billion-value
- JUVE: legal AI as the firm’s operating system (24/08/2026) — https://www.juve.de/markt-und-management/legal-ai-wird-zum-betriebssystem-der-kanzlei/
- Harvey II with Memory (18/08/2026) — https://legaltechnology.com/harvey-ii-launches-with-memory-that-learns-how-individual-lawyers-work/
- Charlotin database, Germany figure (NJW column, 18/09/2026) — https://www.beck-aktuell.de/aus-der-njw/kolumne/b-note-2026-09-18
- Reference case AG Köln, 312 F 130/25 (02/07/2025) — https://legal-tech-verzeichnis.de/legal-tech-nachrichten/anwalt-blamiert-sich-mit-halluziniertem-ki-schriftsatz-gericht-spricht-von-verstoss-gegen-brao/
- Admitted lawyers as of 01/01/2026 (Soldan) — https://www.soldan.de/insights/anwaltsmarkt-2026-immer-weniger-anwaelte-aber-mehr-umsatz/
This article is general information and not legal advice in an individual case. As of 23 September 2026; the figures from the Legal Tech Monitor 2026 come from a sponsor-funded association survey.
Leon Lotz
Leon Lotz is a business lawyer and founder of MusketierSoftware. He combines legal depth with real software craft.
