SNCF builds a private ChatGPT-style assistant and opens it to 100,000 employees

The problem

SNCF counts about 250,000 staff and handles sensitive material: transport plans, maintenance documents and tenders. When ChatGPT appeared, SNCF first let employees use it without restriction, and between 500 and 2,000 staff were using it daily by the end of 2022. The group then chose to build a private version to keep its data inside its own environment.

What they built

SNCF Group GPT is a secure assistant with several spaces. One is a document space that answers questions from more than 70,000 reference documents, for example technical manuals when a technician works on a train axle, with sourced summaries. Another lets office staff upload their own documents for summaries and analysis, which suits tender responses in legal, purchasing, maintenance and sales teams. Retrieval-augmented generation (RAG), where the tool first finds passages in company documents and then writes an answer, powers the document space.

How they did it

An SNCF subsidiary director says protected access opened in September 2023. By February 2024 the tool was in a test phase open to senior executives, with a prompt academy to teach good use. It originally ran on Microsoft's Azure OpenAI service. By 2025 SNCF said its secure environment gave access to OpenAI, Mistral and Anthropic models, and that company data must stay in the SNCF environment. The group has run a wider AI training programme, with 25,000 people trained by mid-2025.

Results

In June 2024, SNCF's deputy chief executive for digital said more than 2,500 employees used the tool daily. In July 2025, the group's digital and AI director said it was open to 100,000 of 160,000 employees in France and that 15,000 used it regularly. He called the gap an area for continued training and change work.