Legal professions
Technical documentation for lawyers, in-house counsel and bailiffs evaluating an AI voice dictation tool with regard to their professional obligations. Last updated: 17 May 2026.
This page describes the characteristics of Fast Dictate's Pro plan.
The technical and contractual characteristics described below (full processing in France, self-hosted AI models, no third-party API calls, DPA available) are specific to the Pro plan. For a complete description of the service's security architecture and GDPR compliance, see the Security & GDPR compliance page.
For a lawyer, voice dictation is a considerable time saver: submissions, procedural pleadings, client interview notes, formal notice letters, legal research. The dictated content, however, falls within the scope of professional confidentiality and the privilege protecting communications between a lawyer and their client, all of which must be taken into account when choosing a technical tool.
A significant portion of AI dictation tools available today rely on public programming interfaces operated by US providers (OpenAI, Anthropic, AssemblyAI, Deepgram, AWS, Azure). The audio stream and transcribed text then transit through infrastructure located outside the European Union. Depending on the firm, this characteristic may raise questions with regard to the applicable professional obligations; it is up to each lawyer or to the head of their local Bar Association to assess its scope.
Fast Dictate Pro offers an alternative approach: an architecture operated entirely in France for dictated professional content, with self-hosted AI models and no calls to a third-party AI API. This page documents that architecture so that each firm can evaluate it against its own requirements.
The references below are provided for information. French lawyers know them better than we do; we cite them only to situate Fast Dictate's technical characteristics within their usual evaluation framework. The official and complete text of these provisions is available on Légifrance, the French official legal portal.
This article punishes the disclosure of secret information by a person who holds it by virtue of their status, profession, function or temporary mission. The full wording, applicable penalties and exceptions (in particular those provided for in Article 226-14 of the same code) appear on Légifrance. The article applies to lawyers under the conditions specified by case law and legal scholarship.
This article covers, under the conditions specified by the official text, the consultations sent by a lawyer to their client, the correspondence exchanged between client and lawyer, correspondence between lawyers (subject to official correspondence), interview notes and case file documents. The materials likely to be dictated to a professional tool frequently fall within this scope.
The RIN, drawn up by the National Bar Council (Conseil National des Barreaux), reiterates the general, absolute and time-unlimited nature of lawyer-client confidentiality. The RIN does not, in itself, prohibit the use of technical writing-assistance tools; it requires the lawyer to exercise the usual vigilance regarding the confidentiality of the processing chain they use. Case-by-case assessment is a matter for the competent head of the local Bar Association.
When a firm or its DPO evaluates an AI voice dictation tool, certain technical questions arise frequently. These questions do not prejudge the compliance of any given tool, they correspond to the points usually examined.
Firms using such tools may wish to consult their local Bar Association head or DPO to assess whether the processing chain is compatible with their own professional obligations. This page makes no such assessment with respect to third-party products.
The architecture of the Fast Dictate Pro plan has been designed to meet the usual requirements of professions subject to a duty of confidentiality. The five characteristics below are verifiable and appear in the Data Processing Agreement available on request.
| Characteristic | Implementation |
|---|---|
| 1. No third-party AI API call | The AI models (Whisper Large V3 for transcription, GPT-OSS-120B for language post-processing) are self-hosted on Scaleway GPUs in France. |
| 2. Audio never retained | Audio recordings are processed in memory for the time strictly necessary for transcription, then destroyed immediately (0-day retention policy). No audio file is persisted on our servers or on those of our inference provider. |
| 3. Transcriptions stored locally | The transcribed text is returned to the application installed on the user's workstation, where it is stored locally. Fast Dictate retains no transcription on the server side. |
| 4. No AI training on customer data | Recordings and transcriptions are never used to train, fine-tune or improve an AI model, internal or third-party. This commitment is contractual through the DPA and imposed on our inference provider. |
| 5. ISO 27001:2022 and HDS-certified infrastructure | The entire technical chain (transcription, post-processing, application back-end) is operated in France on the infrastructure of Scaleway SAS and Clever Cloud SAS, certified ISO 27001:2022 and HDS. Note: HDS certification applies to hosts; Fast Dictate, as a publisher, is not itself an HDS host within the regulatory meaning. |
To these five technical characteristics is added a sixth contractual element: the signing of a Data Processing Agreement (DPA) drafted to meet the requirements of Article 28 of the GDPR, which formalises all the commitments above (see section 6).
Fast Dictate operates at the operating-system level: the transcribed text is inserted directly into any active text field on the workstation, via keyboard injection. No specific application integration is required, the tool needs no dedicated partnership or connector.
In practice, this covers all the environments used in a law firm: Microsoft Word, Outlook, the web browser (including the RPVA interface,Réseau Privé Virtuel des Avocats, the French lawyers' VPN), business case-management software, standard text editors, and any application accepting keyboard input.
Dictation of legal grounds, prayers for relief, observations in reply. The text is written directly in the active field, whether it be Word, the browser (RPVA) or the firm's case-management software.
Immediate write-up, after the meeting, of the factual and legal elements of the case. No audio file is retained once transcription is complete.
Letters to clients, exchanges between colleagues, formal notices. Transcriptions remain on the user's workstation: they are not hosted in a third-party cloud.
Oral summaries of decisions, dictation of case-law notes, drafting of internal memos. Language post-processing is limited to punctuation and formatting, with no rephrasing.
The Data Processing Agreement (DPA) proposed to each Pro plan customer has been drafted to meet the requirements of Article 28 of the GDPR. It specifies, in particular:
The DPA is available on simple request sent to contact@fastdictate.com. Review by the firm's DPO or counsel is encouraged. Once the Pro subscription is taken out, its electronic signature is recorded in the customer area.
Although this page specifically addresses the context of lawyers, the technical and contractual characteristics of the Pro plan may be of interest to other professions subject to a duty of confidentiality:
For the specific needs of a notary's office, a medical practice or a corporate legal department, we can clarify the coverage of the applicable characteristics and the procedures for signing the DPA.
The RIN does not, in itself, prohibit the use of technical writing-assistance tools. The usual practice for a firm is to document the tool's processing chain and assess, where appropriate with the competent head of the local Bar Association or the DPO, whether it is compatible with the applicable professional obligations. This page and Fast Dictate Pro's DPA provide the technical elements necessary for that assessment in respect of the Pro plan.
No. Transcriptions are stored exclusively on the user's workstation. No copy is retained by Fast Dictate or shared between customers. Audio recordings are processed in memory and destroyed immediately after transcription.
An interruption of the transcription service temporarily blocks dictation but does not affect the confidentiality of the chain: no audio is waiting on the server side, and transcriptions already produced remain available locally.
Three ways. (1) Reading the DPA, available on request. (2) Reading the Security & GDPR compliance page, which details the technical architecture and infrastructure certifications. (3) Exercising the annual right of audit provided by the DPA.
The most refined experience remains that of the desktop applications for Windows and macOS, where text is written directly in any active text field. An Android application is available and allows, in the vast majority of cases, direct writing in the text fields of other applications. An iOS application is also available, but subject to the constraints of the Apple ecosystem: direct writing in the text fields of third-party applications is not authorised by the system, and transcription is then done in copy-and-paste mode. All variants rely on the same server architecture (transcription in France, audio not retained, transcription stored locally on the device).
Yes. The technical documentation (this page, the Security page, the DPA) is designed to be readable by a Bar Association president, a data protection officer or a compliance officer. We can supplement these elements with an oral presentation on request.
Reminder. This page describes the technical and contractual characteristics of Fast Dictate's Pro plan. It does not constitute legal advice or a guarantee of compliance with the professional obligations of a regulated profession. Assessment of those obligations rests exclusively with each professional and, where appropriate, with the head of their local Bar Association, their professional order or their DPO.
Fast Dictate's Pro plan is intended for law firms and legal professions wishing to have complete technical documentation on their voice dictation tool. To request the DPA or ask a question, write to us at the address below.