The CSSF and the Banque centrale du Luxembourg published their second thematic review on AI on 16 May 2025, based on a June to August 2024 survey answered by 461 financial institutions, an 86% participation rate. It found that 28% of respondents used AI in production or development, that 61% of the 402 reported use cases involved generative AI, and that 64% of respondents let staff use public GenAI tools while 60% of those had no dedicated GenAI policy. Only 5% of use cases were rated high risk under the AI Act, and the report notes that respondents' ratings reflected perceived risk rather than the Act's actual classification.
OFFICIAL TEXT: cssf.lu ↗ · FINAL · LUXEMBOURG (CSSF / BCL)
| Document | CSSF/BCL second thematic review on AI (2025) — Thematic review on the use of Artificial Intelligence in the Luxembourg financial sector (May 2025) |
| Issued by | Commission de Surveillance du Secteur Financier (CSSF), Luxembourg's financial sector supervisor, with the Banque centrale du Luxembourg (BCL) on joint surveys of AI use |
| Type | Survey |
| Status | Final |
| Published | May 16, 2025 |
| Applies to | Investment firms, authorised investment fund managers, credit institutions, e-money institutions and payment institutions supervised by the CSSF (the survey population). A joint CSSF and Banque centrale du Luxembourg survey report; it sets no requirements |
| Official source | cssf.lu ↗ |
| Use cases | Generative & agentic AI · AI governance (general) · Model risk management · AML / KYC · Fraud detection · Customer-facing chatbots · Credit scoring & underwriting · Third-party & vendor AI |
What are the key points of CSSF/BCL second thematic review on AI (2025)?
- Scope: investment firms, authorised investment fund managers, credit institutions, e-money institutions and payment institutions, more than three times the panel of the 2023 survey.
- Adoption: 28% of all respondents use AI in production or development and 22% are experimenting or plan to within 12 months; among banks, payment and e-money institutions adoption rose to 43% from 30% in the previous survey, with credit institutions at 38%.
- Public GenAI tools: 64% of respondents allow staff access (58% without restriction); 54% of credit institutions block access; among entities that allow access, only 40% have a specific GenAI policy or an internet policy that addresses GenAI.
- Use cases: 402 reported, 54% in production and 92% for internal use only; 61% use GenAI, 30% natural language processing and 28% machine learning; 94% of GenAI use cases rely on large language models and 75% on commercial models.
- Top categories: searching and summarising information (43%), process automation (30%), chatbots and virtual assistants (27%), text generation (27%) and translation (19%); for credit institutions AML/fraud detection fell from first to fifth place.
- AI Act: only 5% of use cases were rated 'High Risk', mainly credit scoring, IRB credit risk models and AML/fraud detection, although the report notes the last two are excluded from the high-risk list in Annex III of the AI Act.
- Trustworthiness: 90% of use cases have a human in the loop (77% in 2023); 45% apply bias prevention or detection measures; good or very good auditability fell to 56% of use cases and explainability to 54%.
- Maturity: 43% of respondents have a formally approved AI policy, 84% have or plan AI training programmes, and 54% have security measures for AI-specific vulnerabilities.
What did CSSF/BCL second thematic review on AI (2025) change for banks?
Compared with 2023 the survey widened to investment firms and fund managers and shifted towards generative AI, which now dominates reported use cases. It records higher adoption and more human oversight but lower self-assessed auditability and explainability, and a first, often inaccurate, self-classification of use cases against the AI Act. The CSSF's 2025 annual report announced a third survey, which opened on 17 November 2026.
What did the 2025 CSSF/BCL review of AI find?
Of 461 respondents, 28% used AI in production or development; 402 use cases were reported, 61% involving generative AI and 92% for internal use only. Human oversight rose to 90% of use cases, but good or very good auditability and explainability fell to 56% and 54%. Only 5% of use cases were rated high risk under the AI Act, often inaccurately.
Do Luxembourg financial firms let staff use ChatGPT and other public GenAI tools?
Of the 461 respondents, 64% allow employees to use public GenAI tools such as ChatGPT, Gemini or Claude (58% without restriction) and 36% deny access. Credit institutions are more restrictive, with 54% blocking access. Among entities that allow access, only 40% have a specific GenAI policy or an internet policy that addresses GenAI.
| Date | Document | Status |
|---|---|---|
| Jul 7, 2026 | CSSF communiqué on frontier AI cyber risk (7 July 2026) — Evolving opportunities and risks in artificial intelligence and its adoption | Final |
| Dec 23, 2024 | Luxembourg bill n° 8476 (AI Act implementation) — Projet de loi n° 8476 portant mise en oeuvre de certaines dispositions du règlement (UE) 2024/1689 (règlement sur l'intelligence artificielle) | Proposed |
| May 3, 2023 | CSSF/BCL thematic review on AI (2023) — Thematic review on the use of Artificial Intelligence in the Luxembourg Financial sector (May 2023) | Final |
| Dec 21, 2018 | CSSF White Paper on Artificial Intelligence (2018) — Artificial Intelligence: Opportunities, risks and recommendations for the financial sector (CSSF white paper) | Final |
| Nov 19, 2026 | BNM AI Feedback Statement — Feedback Statement: Artificial Intelligence (AI) in the Malaysian Financial Sector | Final |
| Nov 19, 2026 | BoE/PRA/FCA effective practices on AI cyber risk (Nov 2026) — Effective practices: Cyber risk and resilience management relating to artificial intelligence (AI) | Final |
Follow every document these regulators publish
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning