Non-disclosure agreements are among the most common contract types that Swiss legal departments and law firms handle every day. That frequency is exactly why they are a natural starting point for AI-assisted NDA review in Switzerland. But not every AI tool is fit for this task – particularly in the Swiss legal context, where data protection rules, professional secrecy obligations, and internal compliance requirements place specific demands on any system used.
This article explains what a meaningful AI-assisted NDA review should deliver, which legal parameters matter in Switzerland, and what a concrete workflow can look like in practice.
Why NDAs in Switzerland require particular care
An NDA – or confidentiality agreement – is a legally binding contract under Swiss civil law. It becomes enforceable once signed by both parties and obligates the receiving party to treat specified information as confidential.
In practice, NDAs appear in many contexts: ahead of M&A transactions, during company valuations, in early-stage business relationships, and when sharing sensitive technical or strategic information. Typical durations range from two to five years, and sometimes longer depending on the subject matter.
What is frequently underestimated: a poorly drafted NDA offers no meaningful protection. Clauses that are too broadly worded may not hold up in court. Clauses that fail to explicitly address certain obligations – such as the deletion of information after the agreement ends – leave gaps open. And an NDA that borrows legal concepts from Anglo-American law without adapting them to Swiss law can be unintentionally flawed.
What NDAs typically cover – and what is often missing
A complete confidentiality agreement generally contains:
a definition of confidential information
the purpose of disclosure
the confidentiality obligation of the receiving party
exceptions (e.g., statutory disclosure duties)
a duration
provisions for penalties or damages in case of breach
What is frequently missing or insufficiently covered: a deletion obligation after the agreement ends, a precise boundary between what counts as confidential and what does not, and liability limitation provisions. These gaps are exactly what a structured NDA review should surface.
AI for NDA review: what it actually means
AI-assisted contract analysis is not the same as automated contract management. An AI-assisted NDA review goes beyond keyword search: it recognizes clauses, evaluates their content against a standard, prioritizes risks, and produces concrete drafting alternatives.
Three functions matter most.
First, clause analysis from the perspective of a specific party. An NDA reads differently depending on whether you are the disclosing party or the receiving party. Useful AI tools account for this perspective when analyzing risks – not in a generic, one-size-fits-all way.
Second, comparison against a standard. The question is not only whether a clause exists, but whether it is sufficiently specified. Is there a liability cap? Is there a deletion obligation? These deviations are the actual weak points.
Third, direct feedback into the document. A list of findings is not particularly useful if the results have to be manually transferred into the document afterward. A tool is practically valuable when it can insert improvement suggestions directly at the right position in a Word document – correctly formatted, without copy-paste.
Data protection and professional secrecy: the Swiss dimension
Before uploading an NDA to any AI tool, one question carries particular weight in Switzerland: where is the data processed?
The new Swiss Federal Data Protection Act (nFADP) requires adequate protection of personal data belonging to clients and contractual partners. On top of that, many professional groups are subject to confidentiality obligations – lawyers under Art. 13 of the Federal Act on the Freedom of Movement for Lawyers (BGFA), banks and insurance companies under their own statutory frameworks. Breach of these obligations carries criminal liability.
Many cloud-based AI solutions route data through servers outside Switzerland, sometimes to the United States. The US Cloud Act gives US authorities the ability to access data held by US companies regardless of where that data is stored. For Swiss law firms or banks processing NDA documents that contain client-related information, this is a real compliance issue – not a theoretical one.
CASUS, a Swiss legal AI platform, processes data exclusively in Switzerland and the EU, transfers no data to the US, and has zero data retention – meaning documents are not stored after processing. There is also no human review of submitted content. These are not marketing claims but technical properties that, for many Swiss users, determine whether a tool is legally permissible to use.
More detail on CASUS's security standards is available on the security page.
How AI-assisted NDA review works with CASUS
CASUS offers two complementary workflows for NDA review:
Risk & Quality Review
The Risk & Quality Review analyzes the NDA from the perspective of a defined party. The output is a structured list of findings, each with an assignment to the affected party, an assessment of relevance, and a severity level (low, medium, high). For each finding, CASUS provides concrete drafting suggestions that can be applied directly into the Word document – without losing formatting.
This is particularly useful when a counterparty has sent over an NDA and the task is to quickly assess which clauses are problematic from one party's perspective. Teams that need to develop clear negotiation positions under time pressure save meaningful work with this workflow.
Benchmark
The Benchmark workflow goes a step further: it compares the NDA against a reference standard – such as an internal playbook or a template NDA based on Swiss best practices. The result shows not just which clauses are present, but evaluates their completeness. Is a deletion obligation missing? Is the liability clause drafted without a cap? Is there no provision covering the return of documents?
CASUS surfaces these deviations as prioritized findings and includes a percentage score reflecting how closely the document matches the standard. Teams that regularly receive and review NDAs can quickly gauge how far an incoming document falls from their internal baseline.
AI Chat for targeted questions
The AI Chat is available for follow-up questions after a structured review. Questions like "What happens in case of breach?" or "How is the termination clause worded?" receive answers that link directly to the relevant passage in the document.
In Agent Mode, the chat can also execute direct changes to the document: inserting clauses, adjusting wording, adding missing sections – all respecting the document's structure and formatting.
When AI adds the most value in NDA review
AI-assisted NDA review delivers the greatest benefit for standardized, recurring contract types and for teams handling high volumes. The more NDAs a legal department or law firm processes each month, the more significant the efficiency gain.
The lead-up to negotiations also becomes more focused: when a tool identifies the key weaknesses of an incoming NDA within minutes and provides drafting alternatives alongside each finding, negotiation teams can enter discussions better prepared and with clearer positions.
AI is less suited for highly unusual or structurally complex confidentiality arrangements – for example, multi-tier joint venture structures involving several disclosure levels, where a comprehensive legal assessment by a qualified lawyer remains necessary.
Practical considerations before getting started
A few points worth clarifying before deploying any AI tool for NDA review:
Is there an internal playbook or template NDA to benchmark against? Without a reference standard, a benchmark tool adds limited value. In that case, a Risk & Quality Review is the better starting point.
Has the tool's data protection compliance been assessed for the Swiss context? Particularly when working with client data or sensitive business information, questions about hosting location and data transfer are not optional – they are legally required due diligence.
Is the output format compatible with existing workflows? Findings that can only be exported as a PDF are of limited use if the document needs to be edited in Word.
Try now
For Swiss law firms and in-house legal teams that review NDAs regularly, CASUS offers a structured workflow: risk analysis from the party's perspective, comparison against internal standards, and direct improvement suggestions in the Word document. Data is processed exclusively in Switzerland and the EU, with no transfer to the US and no data retention. CASUS can be tested free of charge at app.getcasus.com/signup.
FAQ
What is AI-assisted NDA review?
AI-assisted NDA review uses an AI system to automatically assess a non-disclosure agreement for risks, missing clauses, and deviations from a standard. The output is a prioritized list of findings with concrete drafting suggestions – not just a summary of what the contract says.
Which clauses are most commonly flagged in NDA reviews?
In practice, NDAs frequently lack: an explicit deletion obligation after the agreement ends, a clear definition of who qualifies as an authorized recipient, a liability cap in case of breach, provisions covering the return of documents, and a sufficiently precise description of what counts as confidential information.
Is using AI for contract analysis in Switzerland compatible with professional secrecy obligations?
It depends on the tool. The relevant factors are whether the provider processes data on servers outside Switzerland or the EU, whether any human review of submitted content takes place, and whether data is retained after processing. Tools with no data retention, hosting in Switzerland or the EU, and no US data transfer can substantially reduce the risk of a professional secrecy violation.
What is the difference between a Risk Review and a Benchmark for NDAs?
The Risk Review analyzes the NDA from one party's perspective and evaluates existing clauses for risks and weaknesses. The Benchmark compares the document against a reference standard and identifies which standard clauses are missing, incomplete, or divergent. Both workflows complement each other.
Can an AI tool improve the NDA directly in the Word document?
With CASUS, improvement suggestions from the analysis can be applied directly into the Word document – correctly formatted and positioned, without manual copy-paste. The AI Chat in Agent Mode also allows targeted changes based on questions or instructions.
How long do NDAs typically last in Switzerland?
Typical durations are two to five years. In specific areas – such as technical trade secrets or M&A contexts – longer or indefinite confidentiality obligations are also agreed upon. The duration should be explicitly defined in the contract.
Is a verbal NDA valid in Switzerland?
Oral NDAs are technically possible under Swiss civil law but are not advisable for evidentiary reasons. In a dispute, the content of a verbal agreement is very difficult to prove. For any legally significant context, a written NDA should be concluded.
Can AI fully review complex, multi-tier confidentiality structures?
Straightforward and moderately complex NDAs are well suited for AI-assisted review. For highly unusual structures – such as multi-tier joint ventures with several disclosure levels – a comprehensive legal assessment by a qualified lawyer remains advisable. AI does not replace legal judgment; it prepares the analysis in a structured way.
