待翻译：French lawyers are being told to use open source AI models for confidential data

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Back to Articles a]:hidden"> French lawyers are being told to keep confidential data out of ChatGPT, Claude, and Gemini Community Article Published August 9, 2026 Upvote 1 Frank Denis jedisct1 French lawyers have now been told, in unusually blunt terms: do not send confidential client data to Claude, ChatGPT, Gemini, or any other proprietary cloud AI. And even enterprise plans that promise “no training” do not solve the problem. At its General Assembly on 12–13 March 2026, France’s National Bar Council (CNB), which represents French lawyers, made the rule explicit: never disclose client or case data to external generative AI services. Doing so may expose a lawyer to disciplinary or legal sanctions. The CNB rejects the standard cloud AI reassurance. A proprietary provider may promise not to reuse uploaded data, but that claim is generally “neither provable nor verifiable.” In other words, contracts and “no training” assurances are not proof of confidentiality. And data sovereignty is not simply a region selector. The CNB says firms must consider where the data and models are hosted, who controls the servers, and whether those companies are subject to extraterritorial laws—explicitly pointing to US companies. This directly affects Claude for Legal Teams. Anthropic says that commercial data is stored in the US and that traffic may be routed through the US, Europe, Asia, and Australia. So, marketing the service to legal teams does not make the cloud sovereign. The CNB therefore identifies two structural ways to eliminate sovereignty risk: anonymize or pseudonymize the data before it leaves, or install the LLM on the firm’s own systems—for example, by running a local open model. But genuine anonymization is much harder than simply removing names. Amounts, project names, and writing style can re-identify a client when combined. And for workflows such as live meeting transcription, prior pseudonymization may simply be impossible. The practical recommendation for confidential work is local inference: run open-source or open-weight models on machines and infrastructure you control. Keep Claude, ChatGPT, Gemini, and remote APIs for genuinely public or irreversibly anonymized data. And lawyers are not unique in this regard. Doctors, accountants, HR teams, consultants, engineers, researchers, and public bodies also handle legally, contractually, or professionally protected data. They therefore face the same architectural question: should that data leave their systems at all? This is also unlikely to stop at France. The CCBE, which represents Europe’s bars and law societies, already describes on-premises AI as the most secure option for confidentiality. So, expect the same scrutiny across Europe and across professions built on secrecy and trust. If your work touches confidential or sovereignty-sensitive data, stop treating proprietary cloud AI as the default. Instead, run open models locally. And if you sell AI services in Europe, offer a genuinely local or self-hosted mode. A contract is not data sovereignty. French National Bar Council: "even if the provider of a proprietary model states that uploaded data will not be reused, this claim is generally neither provable nor verifiable. The only ways to guarantee sovereignty, isolation and confidentiality are anonymization and local Al with an open-source model." CNB announcement Full guide Le CNB a codifié la conformité IA Community Upload images, audio, and videos by dragging in the text input, pasting, or clicking here. Tap or paste here to upload images · Sign up or log in to comment Upvote 1