The quality of an answer is not enough. A professional tool must above all make it possible to check its sources and protect the firm’s data.
The risk of error exists with every AI. An answer can look convincing while citing a wrong rule or a reference that does not exist. The lawyer must therefore be able to check quickly what the tool states.
Why a general-purpose AI is not always enough
General-purpose assistants are useful for rephrasing a text, summarising an idea or preparing an outline. For a precise legal question, their limit appears when it becomes difficult to know exactly which sources the answer rests on.
In Swiss law, this point is essential. A statute may have changed recently and a new decision may alter the interpretation of a rule.
A well-written answer is therefore not necessarily a correct answer. Before using it in a file, the professional must be able to find the statutory provision, the decision or the document that supports it.
Check the source of every answer
An AI can sometimes invent a reference or attribute to a source content that it does not contain. This is generally called a hallucination.
A reliable legal assistant must make it possible to go back directly to the sources used. For a rule of federal law, the lawyer must for example be able to consult the official text on Fedlex. For a piece of case law, the lawyer must be able to open the decision concerned and check the relevant passage.
If the references are missing, hard to find or too general, the answer requires more checking work.
The aim is not to believe the AI, but to use it to speed up the research while keeping control of the reasoning.
Protect the firm’s data
Legal documents often contain sensitive information: contracts, correspondence with clients, procedural exhibits, business secrets or personal data.
Before using an AI solution, it is necessary to know where the data is stored, who can access it and whether it can be transferred abroad.
The Federal Act on Data Protection (FADP) does not require all data to be hosted in Switzerland, but it does regulate its disclosure abroad. For files covered by professional secrecy, the place of processing remains an important criterion.
It is also necessary to check that the firm’s documents are not used to train the provider’s models and that the data is protected during storage and transmission.
Our article on AI and data protection presents this framework in more detail. The measures stated by Silex are available on the security and professional secrecy page.
Take the file’s internal documents into account
Legal research does not rest only on the statute and the case law. The facts of the file are found in the contracts, emails, internal notes and the other exhibits of the mandate.
A useful assistant must be able to analyse these documents while protecting their confidentiality.
At Silex, this function is offered with Silo. The tool can match the firm’s internal documents with the legal sources used in the analysis.
This approach makes it possible, for example, to take into account the date of a contract, a negotiated clause or a particular event in the file, instead of producing an answer that is too general.
The method is explained in our article on AI-assisted legal analysis.
Choose a transparent provider
For a legal professional, the rules on confidentiality and the use of data must be easy to understand.
The provider should explain clearly where the data is processed, how it is protected, whether it is used to train the models and which persons can access it.
The commercial terms also matter. Prices, usage limits, available storage and any extra costs must be clearly stated.
The more precise this information is, the more the firm can assess the risk before using the tool on real files.
Analysing contracts with AI
Contract analysis is one of the most practical uses of legal AI.
An assistant can carry out a first reading of a contract, find certain clauses and compare several versions of a document.
It can also bring out the differences from an internal template or flag an unusual wording.
The lawyer must, however, understand why a clause deserves attention. A simple flag is not enough: it is always necessary to go back to the contract, analyse the context and determine the legal consequences for the client.
AI speeds up the first reading, but classifying the risk remains a legal task.
Searching case law and preparing a note
AI can also make case-law research easier by allowing a question to be asked in natural language rather than using keywords alone.
It can help to find relevant decisions, summarise their reasoning and prepare a first structure for a note.
The professional must then open the decisions cited, read the important passages in their context and check that they actually concern the question under study.
The necessary checks are set out in our guide on Swiss case law with AI.
Integrating AI into everyday work
A tool is easier to adopt when it integrates with the software already used by the firm.
The integration of Silex into WinLex and SkyLex makes it possible, for example, to start a search without leaving the usual work environment.
The partnership between Silex and MCR Solutions pursues the same aim: to integrate AI-assisted legal research into existing processes.
The time saved concerns mainly preparatory tasks: summarising a file, finding a piece of information, comparing clauses or preparing a first search.
Uses nevertheless vary according to the profession. The solutions by practice page presents the needs of lawyers, notaries, fiduciaries, in-house lawyers, banks and insurers.
The collaboration between Silex and EXPERTsuisse concerns in particular audit, tax and fiduciary professionals. For individuals, see our article on legal AI for individuals.
Human validation remains essential
AI can speed up research and drafting, but it does not replace professional judgment.
A generated answer should never be taken over directly into a procedural exhibit, an opinion or a final contract without checking.
The lawyer remains responsible for the content sent to the client or to an authority. The lawyer must understand the reasoning proposed, check the sources and adapt the result to the context of the file.
The tool must not promise the outcome of a dispute either, or present an answer as a definitive consultation.
For concrete rules on use and review, see our article on how to use a legal chatbot.
Comparing costs
The price must be compared with the functions that are actually useful: quality of the sources, data security, analysis of internal documents, integrations and support.
Silex offers a Solo plan at CHF 149 per month for one user. The Team plan costs CHF 490 per month for up to five users and the Enterprise plan CHF 895 per month for up to ten users.
The differences between the plans concern in particular the number of users, the credits available for certain functions and the storage space. Details appear on the Silex pricing page.
A cheaper tool is not necessarily more economical if the professional has to spend a lot of time correcting its errors or checking its references.
What Silex offers
Silex is designed for research and analysis in Swiss law. The answers are accompanied by sources and Silo makes it possible to work with the firm’s internal documents.
According to Silex, data is hosted in Switzerland and client files are not used to train the models.
The features are presented on the Silex product page.
The private bank Bordier & Cie has adopted Silex for its legal research.
To test the tool on your own files, you can book a demo.
In short
A good legal AI assistant must make it easy to check its sources, protect the firm’s data and take into account the real context of the file.
It must also integrate into everyday work without replacing the professional’s judgment.
AI can speed up research, contract analysis and drafting. Validation and responsibility for the result remain those of the lawyer.
FAQ
Does AI replace the lawyer?
No. It can help to search, analyse and draft, but validation and responsibility for the result remain human.
How can you tell whether the firm’s data is protected?
You need to check the hosting location, access rights, encryption and above all whether the files can be used to train the provider’s models.
How can you limit legal hallucinations?
Use a tool that makes it possible to find the sources of its answers and check the important references before any use in a file.
