AI for lawyers and law firms: what it is good for, and how to keep client secrets
AI for lawyers is mostly a writing and reading assistant — drafting, summarising files, translating, reviewing contracts — and its real risk is confidentiality: whatever you type goes to the company that runs the model. Evanesca is a private AI agent and chat: it runs its models in Europe, keeps no copy of your messages, and signs an Art. 28 DPA with firms. This page sets out what the bar says and what to check, so you can judge any tool, Evanesca included.
Facts checked on 6 Nov 2026
What lawyers use AI for
- Drafting: a first version of a letter, a clause, a memo or a section of submissions, which you then rewrite.
- Summarising: a long judgment, a file, an exchange of emails, reduced to the points that matter.
- Translation: a contract or a decision into another language, to read before a sworn translation.
- Contract review: listing the dates, amounts and obligations in a contract, or comparing two versions.
- Research, with care: explaining a concept or finding leads — never as the source of a citation.
What it does badly is the part that matters most in court: references. A model can produce a case that does not exist, in the right format. The Conseil national des barreaux (CNB), the French national bar council, lists French decisions of December 2025 to February 2026 in which judges asked counsel to check case-law references that did not correspond to published decisions — the Périgueux judicial court, the Orléans administrative court, the Bordeaux administrative court of appeal source source.
The confidentiality problem: client secrets in a consumer chatbot
The CNB describes the mechanism plainly: every request sent to a generative AI transmits data to the company that runs it — in most cases an American one — which keeps and processes it, sometimes to improve its models source. For a lawyer that is not a privacy preference; in France, professional secrecy is a rule of public order:
« Le secret professionnel de l’avocat est d’ordre public. Il est général, absolu et illimité dans le temps. »
A consumer chatbot is built the other way round. On ChatGPT's plans for individuals, as OpenAI's pages read on 6 October 2026, conversations are stored “on OpenAI systems and our trusted service providers’ systems in the US and around the world” until you delete them source, may be used for training unless you opt out source, and may be accessed by authorised staff and service providers source. OpenAI's own advice is: “Please do not enter sensitive information that you would not want reviewed or used” source. Evanesca vs ChatGPT goes through it line by line.
What the bar says about AI for lawyers
Rules differ from one country and one bar to the next; read yours at the source. This page quotes only the French guidance, which we read on 6 November 2026: the CNB's practical guide on ethics and artificial intelligence, first edition, April 2026. Its core rule is strict — the lawyer must never send data about a file or a client to a generative AI:
« Il est impératif pour l’avocat de veiller à ne jamais communiquer des données relatives à ses dossiers ou à ses clients à des intelligences artificielles génératives (IAG). »
- Pseudonymise or anonymise systematically before using a generative AI tool, and limit the context to what the task strictly needs source.
- Check every output: the lawyer must reread, check and validate each generated text before using it or sending it to a client or a court source.
- Prefer, for research, the legal AI tools of legaltechs and legal publishers, whose databases are more reliable source.
- Weigh data sovereignty, in this order: where client data is hosted (France, or the EU), the nationality of the company that owns those servers — excluding companies subject to extraterritorial access, such as American companies under their current law — and the same two questions for the model source.
- Do not rely on terms of use alone: a provider's promise not to reuse data is, in general, neither provable nor verifiable source.
A checklist for law firms choosing an AI tool
- Retention — what is kept after the answer, by the tool and by the company that runs the model, and for how long.
- Training — whether your firm's prompts can train a model, by setting or by design.
- Jurisdiction — which company runs each link (the app, its host, the model) and under which law; a European region of a US company is still a US company.
- Who can read — the provider's staff, the operator's abuse review, sub-processors.
- A data processing agreement under Art. 28 GDPR, with the list of sub-processors and notice before any change.
- Access control — single sign-on and roles, so leavers lose access with their account.
- Logs — what is logged around the content, and who can see it.
- Sources — whether answers cite what they rely on, so a colleague can check them.
The same questions, for any private AI and not only for law firms, are on private AI; why EU hosting is not enough explains the jurisdiction point.
How Evanesca fits a law firm, criterion by criterion
Each row answers one of the CNB's criteria or one line of the checklist, with the facts as they stand; whether they are enough for your practice is your judgement.
| Question | Evanesca |
|---|---|
| Where the model runs | In France by Scaleway and OVHcloud (as a fallback), in Mistral AI's own EU region, in Germany by Verda and AKI.IO and in Switzerland by Infomaniak sovereignty |
| Who runs it | OVHcloud, Scaleway, Mistral AI and Infomaniak are European companies with no US parent, out of the CLOUD Act's direct reach; AKI.IO and Verda do not publish their full shareholding, so Evanesca cannot say that they have no US parent sovereignty |
| Evanesca's own servers | The website, the chat and the agent's loop run at Vercel, a US company, in its Paris region, with no copy of your messages kept there privacy |
| What is kept | Not stored by Evanesca: held in memory while the answer is written; your history stays in your browser; of your messages, only billing data — tokens and cost privacy |
| What the operators keep | OVHcloud keeps “only the data required for billing purposes” source; Scaleway may keep and read requests that cause errors or look abusive, for up to 14 days source; Mistral AI keeps no input or output beyond the time needed to answer, as Evanesca's account has been on zero data retention since 17 November 2026 source, and lets its customers object to training — Evanesca's account objects source source; AKI.IO says prompts and outputs are neither logged nor stored, and keeps usage metadata for 90 days source source; Verda says it stores no request and holds the context in an in-memory inference cache removed within 2 hours source; Infomaniak states that queries are “neither recorded nor used to train models” source |
| Training | Never by Evanesca; Mistral AI lets its customers object, and Evanesca's account objects |
| Who can read | No one at Evanesca; the operator processes the prompt in clear to write the answer — there is no end-to-end encryption |
| Contract and access | For Business workspaces: a signed Art. 28 DPA, the list of sub-processors with notice before any change and the right to object, SSO and roles, spending limits business |
| Record of each answer | A passport under every answer: model, operator, country, law, what was kept passport |
For a whole task rather than a question — “list every deadline and amount in this lease, in a spreadsheet” — Agent mode reads the files of the conversation, writes Word, PDF and Excel documents, never runs code and stops at a cap you set, €2 by default Agent mode · private AI agent. Web search is off by default; switched on, the queries the model writes go to Linkup, a French company web search.
What Evanesca does not do for lawyers
- No legal database and no case-law service of its own: with web search on, it reads public web pages only. For research, a legal publisher's tool is the better choice, as the CNB itself advises.
- It can be wrong, and confidently so — a date, a clause, a reference. Check every output against the source before it leaves the firm.
- No end-to-end encryption: the operator reads the prompt to answer it. Choosing Evanesca changes where and by whom your words are processed; it does not lift the CNB's advice to pseudonymise.
- Its own servers run at a US company, Vercel: if your reading of the CNB's sovereignty criteria excludes any US company on the route, Evanesca does not meet it, and a model installed on the firm's own systems — which the CNB names as one way to remove the risk source — does.
To try it on your own, pseudonymised, material: credit from €5, no subscription required pricing.
Frequently asked questions
What is the best AI for lawyers?
It depends on the task. For legal research, a tool from a legal publisher, with its own database of decisions and commentary, is more reliable — the French bar council advises it. For drafting, summarising, translating or reviewing documents, a general AI assistant works well, provided you check what it keeps, who runs it and under which law.
Can lawyers use ChatGPT?
Many do, for tasks with no client information. On its plans for individuals, OpenAI stores conversations in the US and around the world, may use them for training unless you opt out, and lets authorised staff access them. The French bar council's guide tells lawyers never to send file or client data to a generative AI, and to pseudonymise what they do send.
Is there a private AI for lawyers?
Several tools promise not to keep or train on your prompts; check each one against a written answer on retention, training, jurisdiction, who can read and a data processing agreement. Evanesca runs its models in Europe, keeps no copy of your messages and signs an Art. 28 DPA; its own servers run at Vercel, a US company, in Paris, and the operators that run the models have their own rules.
Can Evanesca do legal research?
Not as a legal database: it has no case-law service of its own, and with web search on it reads public pages only. It can explain, summarise and draft, and it can be wrong, so every reference must be checked at the source.
Does Evanesca sign a DPA with law firms?
Yes. Business workspaces get a signed Art. 28 GDPR data processing agreement, the list of sub-processors with notice before any change and the right to object, single sign-on and roles.
Read next
- The best ChatGPT alternatives, by what you needThe best ChatGPT alternative depends on your need: privacy, European law, free use, research, code or offline. Claude, Lumo, Duck.ai, Vibe and more.
- Private AI agent: what makes an agent private, and how Evanesca's worksA private AI agent carries out tasks without keeping your data. What to check — where it runs, what it keeps, what it can touch — and Evanesca's Agent mode.
- Private AI: what it means, and how to judge onePrivate AI keeps your words from being stored, read or used for training. The three kinds, a checklist to judge any of them, and the main options compared.
- Is there a European ChatGPT? What it means, and the optionsIs there a European ChatGPT? What to check — EU company, servers in Europe, the CLOUD Act — and how Mistral Vibe, Lumo and Evanesca compare.
- European AI: who makes it, who runs it, and how to use itWhat European AI covers: models made in Europe, European operators and data centres, the AI Act — which ones Evanesca runs, and how to use them today.
- Sovereign AI: what the word covers, and how to check a claimSovereign AI means control of the model, the servers, the operator and the law. A checklist to judge any claim, and where Evanesca stands on each layer.