by Ketl
July 22, 2026
7 min read
Legal AI Switzerland: landscape and sovereign solutions
Legal AI Switzerland: legal framework, a 5-question evaluation grid, and a landscape of sovereign solutions for law firms and regulated professions.
TL;DR
Evaluating legal AI from Switzerland raises a different question than an international ranking: where does the data go, and is that destination compatible with professional secrecy (art. 321 SCC), the nLPD and, for financial firms, FINMA oversight? This article starts from that Swiss legal framework, offers a 5-question evaluation grid applicable to any tool, briefly places international solutions through that filter, then details where Ketl stands, a document AI designed and hosted in Switzerland for regulated professions.
Introduction
Searching for legal AI from Switzerland means asking a more precise question than a simple global ranking of the best tools. A Swiss law firm, fiduciary or bank is not just looking for the most capable tool, but one whose use remains compatible with its obligations: professional secrecy, data protection, regulatory oversight. That question structures everything else, and it changes the answer.
At Ketl, we have supported Swiss law firms, fiduciaries and banks since 2019 in adopting document-focused artificial intelligence. This article starts from the Swiss legal framework, offers an evaluation grid applicable to any legal AI tool, then shows where Ketl answers this need directly.
The Swiss framework that structures every evaluation
Three legal texts define what a legal AI tool must satisfy to be used without reservation in Switzerland.
Art. 321 of the Swiss Criminal Code makes breach of professional secrecy a criminal offence. For a lawyer, notary or doctor, transmitting documents covered by professional secrecy to a third-party service whose confidentiality is not guaranteed can amount to disclosure to a third party, with direct criminal consequences.
The nLPD strictly regulates data transfers outside Switzerland and imposes traceability obligations on the processing of personal data, including when that processing runs through an artificial intelligence system.
FINMA oversight, for banks, wealth managers and insurers, requires precise documentation of where data lives and who accesses it, with particular attention paid to external providers and IT outsourcing.
The Heppner ruling made this question urgent
In February 2026, in United States v. Heppner, Federal Judge Jed S. Rakoff ruled that documents generated via a consumer AI tool were not protected by professional privilege, on the ground that transmitting them to a third-party service amounted to disclosure. That decision is American and does not apply directly under Swiss law, but its lesson is transferable: once a document covered by professional secrecy passes through a service whose confidentiality conditions are insufficient, the risk of qualification as third-party disclosure exists under art. 321 SCC as well. Several cantonal bar associations have published guidance on this topic since early 2026.
The 5-question evaluation grid
Before adopting a legal AI tool, regardless of its reputation or ranking position, five questions help verify its compatibility with the Swiss framework.
1. Where is the data physically processed? Hosting advertised as a "Swiss region" or "European region" by a US provider does not answer the question: the vendor's jurisdiction often takes precedence over server location.
2. Does the tool rely on a third-party LLM? If so, which one, and where does that LLM process the data? A tool that transmits identifiable data to a third-party model operated by a foreign entity introduces an additional intermediary into the confidentiality chain.
3. Is the data used to train models? Even anonymised, this question determines whether a confidential document contributes, in some way, to improving a system that also serves other clients.
4. Is the vendor subject to a foreign jurisdiction exposed to the Cloud Act? The US Cloud Act can compel a US-domiciled provider to hand over data to a federal authority, even if that data is physically stored in Europe.
5. Does using the tool preserve professional secrecy in light of rulings like Heppner? This final question synthesises the previous four: given the actual destination of the data, does using the tool expose the firm to a qualification of third-party disclosure?
How to use it: put these five questions in writing to every vendor you evaluate, including Ketl. A vendor that answers these five questions plainly and in writing has generally built its product with this requirement in mind. A vendor that deflects or points to general contractual clauses has probably not built sovereignty in as a structural feature.
International solutions against this filter
The legal AI market includes powerful, widely adopted tools: Claude for Legal, Harvey, Legora for drafting and legal research assistance, Westlaw Edge for AI-augmented case law research. None of these tools was designed with Swiss sovereignty as a starting requirement, which takes nothing away from their performance on their own terrain: reasoning, drafting, research.
| Solution | Origin | Data processing |
|---|---|---|
| Claude for Legal | United States | Cloud, proprietary LLM |
| Harvey | United States | Regional, third-party LLM |
| Legora | Sweden (EU) | EU cloud, third-party LLM |
| Westlaw Edge | United States | US cloud |
Against the 5-question grid, these solutions generally answer transparently on their security certifications, but the question of jurisdiction and Cloud Act exposure remains open for each, insofar as the vendor or its underlying infrastructure falls under US law. For a detailed solution-by-solution analysis, our full comparison of the 10 most searched legal AI solutions covers each tool individually.
Ketl: a document AI designed from Switzerland
Ketl answers a different need from the drafting copilots cited above. Where Harvey or Legora help reason and draft faster, Ketl is a sovereign, AI-augmented DMS (document management system), built for the classification, extraction and document search needs of Swiss law firms, fiduciaries, banks, insurers and international organisations.
Against the 5-question grid, Ketl's answer fits in one line: the entire chain, storage, processing, AI models, is hosted and operated in Switzerland, by a Geneva-based company. No document leaves the territory, including during AI processing. The models used are Ketl's own, developed and trained in Switzerland, roughly 1,000 times lighter than a large generalist model, efficient enough to run on sovereign infrastructure without calling an external service. No third-party American LLM is involved anywhere in the processing chain.
For a law firm subject to art. 321 SCC, this architecture is not a marketing argument: it determines whether the tool can be used without professional liability reservation.
In figures: Ketl processes over 46 million documents across 11 regulated sectors. Clients report a 50 to 80% reduction in archiving time, break-even at an average of 4 months, and ROI between 100 and 350% over four years. Price: CHF 39 to 89 per user per month, all-inclusive, deployment in 4 weeks.
Both logics, drafting copilot and sovereign DMS, can coexist within the same firm: a copilot on anonymised documents for drafting, Ketl for storage, classification and archiving of the originals. What matters is defining, from the outset, which documents may pass through a third-party service and which require fully sovereign processing.
FAQ
Are there legal AI solutions developed in Switzerland?
The market for drafting and legal research copilots (Harvey, Legora, Claude for Legal) remains dominated by American and European, non-Swiss players. Ketl occupies a different category, that of the sovereign, AI-augmented document DMS, with a fully Swiss chain end to end.
Can a Swiss law firm use Harvey or Legora without risk?
It depends on the documents involved. For documents not covered by professional secrecy, or anonymised, use raises fewer difficulties. For documents under art. 321 SCC transmitted without anonymisation to a foreign third-party LLM, the question of third-party disclosure remains open, as the Heppner ruling reminded the profession.
What is the difference between a sovereign legal AI and a legal AI simply hosted in Europe?
European hosting is not enough if the vendor itself, or the underlying LLM provider, is domiciled in the United States and subject to the Cloud Act. Sovereignty requires that the entire chain, company, infrastructure and AI processing, falls under Swiss law.
Does Ketl replace a legal drafting copilot like Harvey?
No. Ketl is a sovereign, AI-augmented DMS, centred on classification, extraction and document search. LLM-assisted drafting falls within tools like Harvey, Legora or Claude for Legal, which address a different need and can coexist with Ketl depending on use.
How much does a sovereign Swiss legal AI like Ketl cost?
Ketl's price sits between CHF 39 and 89 per user per month, all-inclusive: integrations, onboarding, Swiss support and ongoing updates, with deployment in 4 weeks.
Conclusion
Evaluating legal AI from Switzerland starts with a question most international rankings leave aside: where does the data go, and does that destination remain compatible with professional secrecy, the nLPD and Swiss regulatory oversight. The 5-question grid presented here applies to any tool, including Ketl.
The concrete next step: put these five questions in writing to every solution you evaluate, including the one you may already be using. The answers you get will settle the matter faster than any demo, showing whether the tool matches your profession's obligations.
Want to see it on your own files? We have been supporting Swiss law firms and regulated professions in their digital transformation since 2019. Request a demo or contact@ketl.ch
Written following a conversation with James McGill, co-founder of Ketl. Structured and optimised with Ketl AI, our sovereign AI hosted in Switzerland.
Sources:
- United States v. Heppner, S.D.N.Y., February 2026 (Judge Jed S. Rakoff)
- nLPD — Federal Data Protection and Information Commissioner (FDPIC)
- Cloud Act — official text (Congress.gov)