Banque de France Details Five AI Tools in Use, Including an Assistant That Saves Supervisors Two Hours a Week
The Banque de France has set out how it is using artificial intelligence in its daily work, describing five priority tools deployed during 2025 and early 2026 in an article in its Bulletin No. 266, published on October 27. The piece, by Jérôme Coffinet, deputy director of data and analytical services, and Jean-Brieux Delbos, a data process expert in the human resources directorate, was summarised by Revue Banque and ActuIA.
Three of the five are conversational assistants. One helps the supervisors who examine credit institutions with their analysis, another answers staff questions about human resources, and a third supports employees handling requests from members of the public, such as over-indebtedness cases and the right to a bank account. The two remaining tools flag atypical transactions for the fight against money laundering and terrorist financing, and map structured financial products.
The bank reports early results. The supervisors’ assistant saves around two hours per week for each person using it, the HR and front-desk assistants give correct answers more than 95 per cent and 90 per cent of the time, and the structured product analysis reaches 95 per cent reliability. An internal survey in 2025 found that 45 per cent of respondents were using general-purpose chat assistants less than a year after they were rolled out, and the 2026–2028 strategic plan aims to bring AI to every member of staff.
What it means in practice
For a central bank, the figures matter beyond efficiency. Delbos told Revue Banque that being strong on AI is partly a question of credibility, because it supports public and market trust in monetary policy. The institution has also been studying the wider economic effects of the technology, including research showing a drop in young people’s permanent hiring in AI-exposed jobs.
Safeguards sit alongside the rollout. The bank created a Groupe Conseil IA last year to check that its uses comply with the EU AI Act, and Coffinet described AI as playing an ambivalent role in cybersecurity: it can strengthen defences, but criminals can also use it to find weaknesses. Commercial lenders face the same balance, as shown when Crédit Agricole brought Mistral in to build its group AI platform with cybersecurity as a core track.
The bank’s machine learning work is not new. It experimented with machine learning and deep learning between 2015 and 2023, and put some tools into production from 2021 to detect anomalies in banks’ regulatory reporting. What has changed is the spread of generative assistants to ordinary staff. What to watch next is whether the supervisory assistant changes how inspections are run, how the bank measures errors as usage grows, and what it publishes about the AI Act checks.