The Swiss Code of Obligations (CO) has been the legal foundation for contracts in Switzerland for over a century. Today, AI tools - as used on platforms like CASUS - are changing how contract law is applied in practice. What the CO actually requires, where AI can help, and where human judgment remains indispensable are questions that matter equally to law firms and in-house legal teams.
This article goes beyond a general overview. It describes typical checkpoints in Swiss contract work, names the relevant statutory provisions, and shows which clause types AI can reliably handle - and which it cannot.
The CO as the basis of Swiss contract law
The Code of Obligations governs the formation, content, and termination of contractual relationships in Switzerland. Most of it applies as default law: the parties can agree to deviating rules. Where a contract is silent, the CO steps in as a statutory fallback.
That has significant practical consequences. If a contract for work contains no provision on defect notification, CO Art. 367 ff. applies; for immovable works, the notice period has been 60 days since 1 January 2026, and a shorter period cannot be agreed (Art. 367 para. 1bis CO). If a sales contract is silent on warranty, CO Art. 197 ff. takes effect. For contract interpretation - when the wording of a clause is disputed - CO Art. 18 is the reference point: it makes the actual common intent of the parties the decisive factor, not merely the literal text. The Federal Supreme Court confirmed in BGE 144 III 93 that normative interpretation based on the principle of good faith only applies where an actual common intent of the parties cannot be established.
Leaving out a clause means accepting the statutory default. Sometimes that is deliberate; often it is an oversight. This is one of the clearest areas where AI adds value in contract work: identifying gaps before they become problems.
Where AI fits into Swiss contract law practice
AI-assisted contract analysis is not a substitute for legal expertise. It is a tool that handles certain tasks faster and more systematically than a person working alone.
Identifying gaps and deviations
A common problem in contract work: a contract includes a liability clause but no liability cap. Or there is no provision for data protection responsibility. Finding these gaps manually takes time and requires good recall of standard requirements.
CASUS addresses this with the Benchmark workflow. It checks a document against an internal playbook or established industry standards - for SPA, NDA, or DPA contracts, for example. The system shows which clauses are missing, incomplete, or deviating from the standard, and provides a concrete suggestion for each finding, including the option to insert a suitable clause directly at the right place in the Word document. The benchmark output also includes a percentage match score against the reference standard.
Typical checkpoints in Swiss NDAs are whether a limitation of liability is provided for and whether the confidentiality obligation includes a duty to return or delete information after expiry. For SPAs, this includes whether a MAC provision (Material Adverse Change) is wanted; in Swiss transactions it is often deliberately omitted, while in international deals it is frequently expected. Benchmark makes such gaps against your own playbook visible immediately.
Structured risk prioritisation
A contract with twenty findings is hard to work through without a clear sense of what to address first. The CASUS Risk & Quality Review assigns each finding to a party and prioritises by severity: low, medium, or high. This makes it possible to identify negotiation potential quickly and manage work in a targeted way.
The review itself takes 1 to 3 minutes per contract and returns around 20 findings on average. Time shifts from searching to deciding: instead of combing the contract clause by clause for risks, the lawyer works through prioritised findings and applies suitable wording directly in Word.
Checking consistency and language
Long contracts are prone to subtle errors: a term is capitalised inconsistently; a cross-reference points to a section that does not exist; a placeholder was overlooked. The CASUS Proofread function checks spelling, grammar, terminology consistency, and structural elements such as numbering, cross-references, and annexes - without altering the legal meaning of any clause. Swiss spelling conventions (ss instead of ß, consistent party names) are applied throughout.
The regulatory framework: AI and law in Switzerland
Switzerland currently has no specific AI legislation. On 12 February 2025, the Federal Council mandated the Federal Department of Justice and Police (FDJP) to prepare a consultation draft for AI regulation by the end of 2026. Planned areas of regulation include transparency, data protection, non-discrimination, and oversight. The new rules are intended in particular to enable Switzerland to ratify the Council of Europe Framework Convention on Artificial Intelligence, which it signed on 27 March 2025.
For law firms, the existing professional rules apply in parallel: professional secrecy (Art. 321 Swiss Criminal Code, Art. 13 BGFA) and the duty to practise with care and diligence (Art. 12 lit. a BGFA). AI tools can be used in line with these duties if quality control remains with the lawyer and client data is passed to AI providers only under controlled, contractually secured conditions.
What the revised FADP specifically requires
The revised Swiss Federal Act on Data Protection (revDSG/nFADP), in force since 1 September 2023, is relevant to AI use in contract work in several respects. Art. 6 revDSG codifies the principles governing data processing: lawfulness, good faith, proportionality, and purpose limitation. Anyone feeding contract data containing personal data into an AI system must ensure that processing meets the principles of Art. 6 revDSG.
Art. 5 lit. c revDSG defines sensitive personal data, including health data, biometric data, and data on administrative or criminal proceedings. These categories appear in contracts regularly - in HR agreements, M&A documents, and medical practice transfer agreements. Elevated data security requirements apply.
Where the GDPR applies to the processing (Art. 3 GDPR), Art. 28 GDPR also governs processor arrangements. A DPA without the required processor clauses is then not just incomplete but non-compliant.
CASUS hosts data exclusively in Switzerland and the EU and transfers no data to the US, not even via subcontractors. User data is not retained after a session and is not used for training (Zero Data Retention), and there is no manual access by third parties (No Human Review). Legally, Art. 6 FADP requires lawful, proportionate and purpose-bound processing, and under Art. 9 FADP firms must make sure that their provider guarantees data security. CASUS goes beyond this minimum: with US-hosted tools, firms have to assess and contractually secure third-country transfers, foreign authority access and retention periods themselves; with CASUS, these issues do not arise. This makes use in client mandate contexts considerably more straightforward.
AI-assisted legal research on CO questions
Beyond contract analysis, AI can support legal research on CO-related questions. The CASUS Legal Research module provides access to over 8 million legal documents from six countries, including case law and legislation from Switzerland, Germany and Austria. Relevant reasoning sections from decisions are displayed directly within answers - no need to navigate to the original source.
For Swiss law, CASUS also searches open-access commentary from Onlinekommentar and company information from ZEFIX, the Swiss Central Business Name Index. View the current source overview.
This enables structured first assessments on questions such as: How does case law treat limitation-of-liability clauses in general terms? What are the requirements for extraordinary termination under CO Art. 337? Under the established case law of the Federal Supreme Court, an important reason within the meaning of CO Art. 337 exists when, in good faith, the terminating party can no longer be expected to continue the employment relationship. Such leading decisions can be retrieved directly in Legal Research mode and used as the basis for clause drafting.
The outputs are source-based and traceable - not generated answers without legal grounding. They can be used directly in further work: as a rationale for a clause, as an internal memo, or as the basis for changes via Agent Mode in the chat.
Key CO clauses: what AI can reliably spot - and what it cannot
Not all clause types are equally suited to automated review. The table below distinguishes between default and mandatory CO provisions and shows where AI is reliable and where human judgment is essential.
Clause type | CO basis | Mandatory / default | AI reliability |
|---|---|---|---|
Exclusion of liability for intent/gross negligence | Art. 100 para. 1 CO (auxiliaries: Art. 101 paras 2 and 3 CO) | Mandatory (Art. 100 para. 1 CO); liability for auxiliaries can be excluded within the limits of Art. 101 para. 3 CO | High: clause can be checked for violation |
Liability cap | Art. 100 CO; for consumer general terms also Art. 8 UCA | Default | High: absence reliably flagged |
Ordinary termination (tenancy) | Art. 266 ff. CO | Partly mandatory | Medium: notice periods checkable, cantonal conciliation specifics not |
Termination with immediate effect (employment contract) | Art. 337 CO | Mandatory (paras 1 and 2, Art. 361 CO) | Medium: prerequisites identifiable, individual assessment not |
Warranty in sales contracts | Art. 197 ff. CO | Default; limits in Art. 199, Art. 210 para. 4 and (real estate) Art. 219a CO | High: missing notice periods and exclusions are recognised |
MAC definition (M&A) | No specific CO basis | Individually drafted | Low: absence can be flagged, drafting remains the lawyer's task |
Data protection responsibility (DPA) | Art. 9 FADP; Art. 28 GDPR (where applicable) | Mandatory (processing on behalf; minimum content under Art. 28(3) GDPR where the GDPR applies) | High: presence and completeness checkable |
Contract interpretation clause | Art. 18 CO | Default | Low: interpretation questions require individual assessment |
This distinction has practical consequences. For mandatory provisions, AI reliably detects violations - a clause that violates CO Art. 100 para. 1 is void regardless of what the parties agreed. The final legal assessment, for example of liability for auxiliaries (Art. 101 CO) or of general terms and conditions, stays in the lawyer's hands. For highly individualised clauses such as MAC definitions or governing-law/interpretation rules, the quality of AI analysis depends directly on the quality of the playbook behind it.
Limits of AI in contract law
AI tools are well suited to recognising structured patterns, comparing clauses against standards, and checking formal consistency. For complex transactions, they hit specific limits that are worth naming precisely.
First: without a clear playbook or defined minimum positions, AI produces a list of findings without any ability to assess which of them are genuinely critical to the negotiation. A missing liability cap in a supply contract worth CHF 50,000 is a different risk from the same gap in an SPA worth CHF 50 million. That contextualisation is the lawyer's job.
Second: in heavily negotiated bespoke contracts - for example, joint ventures between two groups each with their own legal team - the contract document often captures only part of the commercial understanding. Side letters, oral agreements, and industry practice are absent from the text but very much present in reality. AI sees only the document.
Third: cantonal specifics are frequently underestimated by AI systems. Tenancy law with cantonal conciliation authority procedures, property transfer taxes in cantons such as Bern or Geneva, and differences in cantonal court and authority organisation and practice - these dimensions are invisible in the contract text but central to the legal assessment.
A typical checkpoint in NDA reviews is the clause requiring the return or destruction of confidential information after contract expiry. Where personal data is involved, the statutory duty to destroy or anonymise data that is no longer needed (Art. 6 para. 4 FADP) applies in any event; the clause sets out this duty contractually and also covers confidential information that is not personal data. AI reliably flags when this clause is missing. Whether to require a written destruction confirmation is a judgment call for the lawyer.
Decision frame: when AI adds value in Swiss contract work
The following breakdown helps assess where AI-assisted contract analysis delivers clear value and where preparation is needed first.
AI delivers high value when:
The document corresponds to a known contract type (NDA, SPA, DPA, construction contract) and a playbook is in place
The goal is identifying gaps, not assessing negotiation positions
Multiple documents need to be reviewed in parallel (due diligence situations)
Formal consistency, cross-references, and terminology need to be checked
Legal research on a specific CO question needs to be structured quickly
AI needs preparation when:
The contract type is highly individualised and no reference standard exists
The commercial logic of the deal is not captured in the document
Cantonal specifics (conciliation authorities, property transfer taxes, local authority practice) are relevant
AI cannot replace:
The weighing of legal risk against commercial pragmatism
The judgment of what is critical to negotiate versus what is acceptable
Information outside the document (side letters, oral agreements, industry practice)
CASUS for everyday contract work under Swiss law
Legal teams working with Swiss contracts who want a structured approach to AI can use CASUS directly in Microsoft Word or via the web app. The platform covers the full document lifecycle: from Benchmark comparison against an internal playbook, through Risk & Quality Review, to Proofread before sending.
For due diligence situations, the AI Data Room is available to analyse dozens or hundreds of documents in parallel and extract clauses - on liability, IP, termination, or any defined topic - into a structured table.
CASUS can be tested free of charge at app.getcasus.com/signup. Further details on data protection and hosting are available at /security. An overview of all modules is at /product.
FAQ
What does the Swiss CO govern in contract law?
The Code of Obligations (CO) governs the formation, content, and termination of contractual obligations in Switzerland. Most of it applies as default law, meaning parties can agree to deviating rules. Where a contract is silent, the statutory CO provision applies as a fallback. Contract interpretation follows CO Art. 18, which makes the actual common intent of the parties the primary reference.
Can AI apply Swiss contract law independently?
AI tools can identify clauses, compare them against standards, and flag gaps. The legal assessment of whether a clause is appropriate in a given situation - and what negotiation strategy makes sense - remains the responsibility of qualified lawyers. For mandatory CO provisions, such as the prohibition on excluding liability for intent and gross negligence in advance under CO Art. 100 para. 1, AI reliably detects violations; the final assessment stays in the lawyer's hands. For highly individualised clauses, the quality of the analysis depends directly on the playbook in place.
Is using AI in Swiss contract work legally permitted?
Switzerland has no specific AI law. The Federal Council mandated the FDJP on 12 February 2025 to prepare a consultation draft by end of 2026. Until then, AI use is permitted under existing law, provided data protection law, professional secrecy (Art. 321 Swiss Criminal Code, Art. 13 BGFA) and professional duties of care (Art. 12 lit. a BGFA) are observed. Lawyers remain responsible for the quality of their work product, so quality control must stay with the human professional.
Which CO clauses are most relevant for AI analysis?
The clauses most commonly reviewed are liability provisions (CO Art. 100/101), termination rules (CO Art. 266 ff., Art. 337), warranty provisions (CO Art. 197 ff.), and data protection clauses in the context of Art. 9 FADP and Art. 28 GDPR. Benchmark also shows whether NDAs include a deletion obligation after expiry and whether SPAs contain a MAC provision.
What does the revDSG specifically require for AI use?
Art. 6 revDSG requires that personal data be processed lawfully, in good faith, and proportionately. Anyone feeding contract data containing personal data into an AI system must meet these principles. Art. 5 lit. c revDSG defines sensitive personal data - including health data and biometric data - which triggers elevated protection requirements. Where the GDPR applies (Art. 3 GDPR), Art. 28 GDPR governs processor arrangements.
How does contract analysis differ from contract management?
Contract analysis covers the substantive review of a document: risk assessment, clause identification, and recommendations. Contract management covers the administration of contracts over their lifecycle - deadlines, renewals, filing, and reporting. CASUS primarily supports the analysis and review phases, not lifecycle-based contract management.
How does CASUS protect data when using AI?
CASUS hosts data exclusively in Switzerland and the EU, transfers no data to the US (not even via subcontractors), does not retain user data after a session or use it for training (Zero Data Retention), and allows no manual access by third parties (No Human Review). Art. 6 FADP (lawfulness, proportionality, purpose limitation) and Art. 9 FADP (processors) set the legal minimum; CASUS goes further, so firms do not need to assess third-country transfers, foreign authority access or retention periods. This makes use in client mandate contexts considerably more straightforward.
What is the difference between a benchmark and a risk review?
A benchmark checks whether a contract meets a defined standard - for example, whether all standard clauses are present and complete - and outputs a percentage match score. A risk review analyses the specific risks and weaknesses of a particular document from the perspective of one contracting party, prioritises by severity, and provides concrete drafting options.
Can AI help with research on CO questions?
Yes. The CASUS Legal Research module provides access to over 8 million legal documents from six countries, including case law and legislation from Switzerland, Germany and Austria. Relevant reasoning sections are displayed directly in the chat. For example, the Federal Supreme Court's case law on the requirements for extraordinary termination under CO Art. 337 can be retrieved quickly and used as the basis for clause drafting. Results are source-based, structured, and directly usable in further work.
