The main risk appears when well-written texts are used without checking the facts, sources or citations.
In law, a fluent answer is not enough. Every important statement must be traceable to a real source and checked by a lawyer.
What is “AI slop”?
The term “AI slop” refers to content produced quickly by artificial intelligence, often well presented but thin on substance or insufficiently verified.
In the legal field, the problem becomes serious when these texts contain invented references, non-existent decisions or rules that have been misinterpreted.
The time saved at the drafting stage can then be lost during verification, because every citation must be checked before it is used.
For individuals who use these tools themselves, our guide to legal AI for individuals sets out the main precautions.
Why AI can invent citations
Language models produce text by predicting the most probable words in a given context.
If they are not working from reliable legal sources, they can generate a reference that looks credible without it actually existing.
A fabricated decision can thus take the name of a court, a plausible case number or a real legal principle while being entirely invented.
The risk increases when a user reuses the answer without checking the sources.
Working from checked sources
To limit errors, AI-assisted legal research must remain linked to sources that can be opened and verified.
The tool can draw on official legal databases, case law, legal literature or the firm’s internal documents.
The professional must then be able to open the source used and check that it really supports the statement produced.
This method makes it possible to use AI as a research assistant rather than as a source of authority.
Protecting the data in the file as well
The reliability of the sources is not the only issue. Documents sent to an AI can contain confidential information.
Before using a tool on a real file, it is necessary to check where the data is stored, who can access it and whether it is used to train the models.
The Federal Act on Data Protection (FADP) does not require all data to be hosted in Switzerland, but it does regulate its disclosure abroad.
The rules are set out in detail in our article on AI and data protection. The measures stated by Silex appear on the security and professional secrecy page.
The courts are watching AI-generated citations more closely
Several foreign courts have already noted filings containing legal references that did not exist or could not be verified.
In Germany, decisions handed down in 2025 have flagged citations that appeared to have been generated by AI in court submissions.
These situations show that using an AI tool does not relieve the professional of the duty to check the sources before filing a document.
In family law, where files are often especially sensitive, our article on AI and family law sets out the uses and limits of these tools.
The signatory remains responsible for the document
When a lawyer files a written submission, they remain responsible for its content, even if part of the text was prepared with AI.
An invented citation, a misapplied rule or a factual error can weaken the file and damage the professional’s credibility.
AI cannot serve as a justification for a lack of verification.
The risks linked to the everyday use of these tools are also discussed in our article on everyday legal AI and the law.
The real cost of poorly checked content
A quickly produced answer brings no gain if it then requires lengthy corrections.
A bad citation can force the team to redo part of the brief, check new sources or correct an analysis already sent out.
There is also a reputational risk. A firm that regularly sends incorrect references can lose the trust of its clients and partners.
The real time saving therefore comes from a tool that also makes verification easier.
Preserving the credibility of the profession
The spread of unverified legal content can make it harder to distinguish a reliable source from a merely convincing text.
Professionals have an interest in adopting clear internal rules: use verifiable sources, check important references and train teams on the limits of generative models.
Our guide on how to choose a legal AI assistant sets out the main criteria to examine.
The form of the text also matters. Overly standardised or artificial drafting can harm the credibility of the document, as we explain in our article on the artificial footprint in legal writing.
How to check a source before citing it
A legal citation must be retrievable from an official or recognised source.
When AI mentions a statute, a decision or a work, the lawyer must open the reference, check that it exists and read the passage in its context.
It is also necessary to check that the reasoning actually matches the source cited. A reference can be genuine and still be misinterpreted.
This check remains essential before any use in an opinion, a contract or a court filing.
Human analysis remains essential
AI can search, summarise and organise large volumes of information.
It does not, however, know everything at stake in the file, the client’s objectives or the strategic choices that have to be made.
The lawyer must place the results back in their context, discard unhelpful suggestions and complete the analysis where necessary.
AI speeds up the preparatory work. The decision and the responsibility remain human.
What Silex offers
Silex is designed to work from Swiss legal sources and the firm’s internal documents.
With Silo, professionals can analyse their own documents and link them to the sources used in the research.
According to Silex, the answers point to verifiable sources, data is hosted in Switzerland by Infomaniak and client files are not used to train the models.
The tool can also be integrated into certain existing work environments. The partnership between Silex and MCR Solutions and the integration of Silex into WinLex and SkyLex present this approach.
The private bank Bordier & Cie also uses Silex for its legal research.
The plans are presented on the Silex pricing page, the features on the Silex product page and the various uses on the solutions by practice page.
In short
AI-generated content must never be used solely because it looks credible.
Before filing or sending a document, check the statutes, decisions and other references cited. Also make sure that the file data has been processed in a suitable environment.
A useful legal AI must make this verification easier, not harder.
To test Silex on your own files, you can book a demo.
FAQ
Can a general-purpose AI be used on a legal file?
Yes for some preparatory tasks, but the sources and the data-processing conditions must be checked before the result is used in a real file.
Who is responsible if AI produces a bad citation?
The professional who uses and transmits the document remains responsible for checking the final content.
