# Regulated professions do not need spectacular AI. They need responsible AI. > Reference HTML page: https://leaid.ai/en/articles/ia-responsable-professions-reglementees > Language: en > Author: Mamadou Waggeh, Founder of Leaid > Date: 2026-07-01 Regulated professions are not hostile to innovation. They are simply bound by particular obligations: independence, professional secrecy, competence, responsibility, traceability and the protection of individuals. AI should therefore not be conceived as a break with these principles, but as an infrastructure compatible with them. ## Introduction Debates on AI in regulated professions often swing between two extremes. On one side, the substitution narrative: machines would soon replace lawyers, in-house counsel, chartered accountants, doctors, notaries and public officials. On the other, the refusal narrative: AI would be too risky, too opaque, too far removed from professional values to be integrated seriously. Both positions miss the essential point. Regulated professions are not defined solely by the tasks they perform. They are defined by the responsibilities they assume. A lawyer does not merely draft a document: they advise, defend, exercise judgement, assume liability and protect their client’s confidentiality. A doctor does not merely read a result: they care for a person. A chartered accountant does not merely produce figures: they guarantee reliable information. A public official does not merely process a request: they take part in a public-interest mission. AI can transform tasks. It must not dissolve responsibilities. It is on this distinction that a French doctrine for the adoption of technology by regulated professions must be built. ## 1. The transformation of work: from production to validation Generative AI is already changing the mechanics of knowledge work. Historically, legal professionals, and the advisory professions more broadly, devoted a significant share of their time to analysing, producing and then delivering. They collected information, characterised it, drafted a deliverable and then passed it on to the client, the judge, the authorities or the organisation. With AI, part of the preparatory analysis and initial production can be accelerated. The tool can propose a summary, spot inconsistencies, extract data, formulate an outline, generate a draft clause or produce a report. The core of the profession then shifts towards [verification, validation and delivery](/en/articles/ia-juridique-maitrise-du-workflow). This shift does not necessarily diminish the professional’s value. It makes it more visible: knowing what is false, incomplete, wrongly characterised, dangerous or unsuitable becomes central. This change is already visible within in-house legal departments and compliance functions. In-house counsel use AI to save time on basic tasks, but must check the results, document their judgement calls, test for bias, keep records and explain decisions. The professional becomes at once user, reviewer, trainer and sometimes trainer of the AI itself. ## 2. Responsible AI begins with human responsibility The first rule is simple: the professional cannot hide behind AI. A deliverable produced with the help of a model remains a professional deliverable. A fabricated citation in a written submission, an erroneous clause in a contract, a misclassified risk, a discriminatory decision or a breach of confidentiality do not become acceptable because a tool generated them. The sanctions imposed on lawyers who filed submissions containing hallucinations are a reminder of this obvious point. Specialised legal tools may hallucinate less than general-purpose models, but they do not abolish the duty to verify. In regulated professions, the standard cannot be “the tool is generally reliable”. It must be “the professional has checked everything for which they are liable”. Responsible AI therefore rests first and foremost on human discipline. Sources must be checked, reasoning tested, useful prompts documented where necessary, validations recorded, [roles organised and teams trained](/en/expertises#adoption-ia). Responsibility does not disappear. It shifts to the design of the control process. ## 3. The risks specific to regulated professions AI-related risks are not uniform. They depend on the profession, the data processed, the people concerned and the effect produced. In regulated professions, several risks must be distinguished. - **Confidentiality risk**: sensitive data, or data covered by professional secrecy, entered into a tool that is insufficiently controlled. - **Hallucination risk**: the production of references, analyses or facts that do not exist, but are presented with confidence. - **Over-delegation risk**: accepting a model’s output without understanding or checking it. - **Bias risk**: reproducing or amplifying discrimination in analysis, prioritisation or decision-making. - **Traceability risk**: the inability to explain how a conclusion was reached. - **Dependency risk**: loss of skills, vendor lock-in, inability to switch tools. - **Dehumanisation risk**: replacing a professional relationship that requires listening, nuance and trust with an automated interaction. These risks do not justify inaction. They justify method. Abstaining is not a sustainable strategy, because clients, users, competitors and staff will gradually adopt these tools. But ungoverned use is just as dangerous. ## 4. Training becomes a professional duty One of the most underestimated issues is training. Using AI is not simply a matter of knowing how to write a prompt. One needs to understand the limits of models, the risks of persuasion, hallucinations, tokenisation, confidentiality, bias, validation chains, the differences between models, contractual issues and how to evaluate an answer. Younger professionals are often more digitally fluent, but that fluency is not enough. More experienced professionals have indispensable professional judgement, but may be less familiar with the tools. AI can therefore create a new intergenerational bond if it is conceived as a space for mutual learning. Juniors help with technical adoption. Seniors pass on method, caution, reasoning and how to relate to clients. Care must also be taken not to destroy the mechanisms of professional training. If organisations eliminate entry-level tasks too quickly, they risk depriving young professionals of learning by doing. A lawyer learns by drafting, rereading, making mistakes and being corrected. An AI that automates basic tasks without rethinking training can weaken the next generation of professionals. ## 5. A doctrine in six principles A French doctrine on AI in regulated professions could rest on six simple principles. - **Principle of purpose**: AI must meet an identified professional need, and not be deployed because it is available. - **Principle of proportionality**: the more sensitive the data and the consequences, the stronger the safeguards must be. - **Principle of responsibility**: every use is the responsibility of a clearly identified professional or organisation. - **Principle of traceability**: it must be possible to document, audit and explain significant uses. - **Principle of practical sovereignty**: supplier choices must take account of data protection, reversibility, support and strategic dependency. - **Principle of continuing training**: AI skills must be built into professional obligations, initial training and development plans. These principles are not intended to slow down innovation. They are intended to make it compatible with the obligations on which [trust in regulated professions is founded](/en/articles/ia-generative-secteurs-regules-confiance). ## 6. The democratic and institutional stakes AI in regulated professions is not merely a matter of productivity. It touches on access to justice, the quality of care, trust in institutions, consumer protection, economic security and equal treatment. The debates on deepfakes, copyright, the content used to train models, the algorithmic filtering of political content and proposals to scan encrypted communications show how quickly technology spills over into democratic issues. Regulated professions therefore have a particular role to play. They can become counterweights, mediators and standard-setters. Lawyers can help define the rights and remedies associated with AI. In-house counsel can turn compliance into a competitive advantage. Notaries, judicial officers (commissaires de justice), chartered accountants, doctors and public officials can help develop uses that protect people as much as they improve processes. France has an opportunity here: to champion an approach to professional AI that is neither technophobic nor naive. An approach that embraces innovation, but refuses to treat data, responsibilities and people as mere optimisation variables. ## Conclusion Regulated professions do not need spectacular AI. They need AI that is reliable, explainable, governed and compatible with their obligations. The question is not whether AI will soon do professionals’ work. The question is how professionals will do their work better with AI, without giving up what justifies their role. Responsible adoption will be no less ambitious than rapid adoption. It will simply be more sustainable. In markets built on trust, innovation does not win when it impresses. It wins when it can be answered for. --- Markdown version of https://leaid.ai/en/articles/ia-responsable-professions-reglementees — Leaid · https://leaid.ai · https://leaid.ai/llms.txt