Operational resilience and cyber risk

FINMA Circular 2023/1

AI systems are part of the information and communication technology of an institution, and their failure or manipulation is an operational risk. FINMA Circular 2023/1 “Operational risks and resilience – banks”, published in December 2022, entered into force on 1 January 2024. It incorporates the principles on operational risk and operational resilience published by the Basel Committee on Banking Supervision in March 2021. The circular contains specific sections on the management of ICT risks, cyber risks and critical data, and it replaced the business continuity recommendations of the Swiss Bankers Association that FINMA had previously recognised. The requirements on operational resilience were subject to a transitional period that ended on 1 January 2026.¹

Operational resilience means that a bank must be able to continue its critical functions through severe disruptions within a defined tolerance. Once AI tools are used in critical processes such as payment monitoring or client onboarding, the institution must be able to operate those processes if the model fails, produces unreliable results or becomes unavailable because a provider suffers an outage. In practice this requires fallback procedures, documented manual processes and tests of disruption scenarios. The category of critical data introduced by the circular also shapes AI projects. Training or operating a model on critical data requires the institution to apply the circular’s controls on confidentiality, integrity and availability to the AI environment.

Cyber reporting to FINMA

Cyber risk is closely linked. FINMA requires supervised institutions to report significant cyber attacks on critical functions under Article 29 paragraph 2 of the Financial Market Supervision Act. Guidance 05/2020 established a two-stage procedure, with an initial notification within 24 hours and a full report within 72 hours. Guidance 03/2024 summarised FINMA’s observations from cyber supervision, clarified the earlier guidance and called for scenario-based cyber exercises. FINMA counts cyber risks among the top risks of the Swiss financial centre.²

AI changes the threat landscape in two directions. Attackers use generative models to produce convincing phishing messages, forged documents and synthetic voices that imitate executives or clients. The Swiss Bankers Association lists deepfakes and fake executive communications among the principal risks of generative AI, as noted above. Institutions that deploy AI systems also create new points of attack, such as prompt injection against chatbots, poisoning of training data and the theft of models. The acquisition of the Zurich company Lakera by Check Point, discussed in section III, shows that the defence of AI systems has become a field of specialisation in its own right.

The NCSC reporting obligation for critical infrastructure

Since 1 April 2025, operators of critical infrastructure have been required by the amended Information Security Act and the Cybersecurity Ordinance to report cyber attacks to the National Cyber Security Centre (NCSC) within 24 hours of discovery. The obligation covers attacks that jeopardise the functioning of the infrastructure, lead to manipulation or leakage of information or involve blackmail, threats or coercion. An initial report may be completed within 14 days. Sanctions for failure to report could be imposed only after a transitional period ending on 1 October 2025.³ The legal basis lies in Articles 74a and following of the Information Security Act.⁴

The Information Security Act includes banks, insurers and financial market infrastructures among the organisations subject to the reporting duty, alongside energy suppliers, transport companies and public administrations.⁵ For supervised financial institutions this creates two parallel reporting lines, one to FINMA and one to the NCSC. Although the two duties overlap in their subject matter, they rest on different legal bases and pursue different purposes. The prudential supervisor assesses the institution’s resilience, whereas the NCSC builds a national picture of the threat. For AI-related incidents, such as a successful attack on a customer-facing chatbot that leads to the leakage of client data, an institution may also have to notify the FDPIC under the data breach provisions of the FADP. In such a case the institution would have to address the question of banking secrecy with its clients at the same time.

Notes

1. FINMA, “FINMA publishes Circular ‘Operational risks and resilience – banks’”, 13 December 2022, https://www.finma.ch/en/news/2022/12/20221213-mm-anh-rs-op-risks/. ↩
2. FINMA, “New cyber supervisory approach and guidance” (dossier cyber risks), https://www.finma.ch/en/documentation/dossier/dossier-cyberrisiken/new-cyber-supervisory-approach-and-guidance/ (accessed September 2026). ↩
3. National Cyber Security Centre, “Obligation to report cyberattacks on critical infrastructure comes into force on 1 April”, https://www.bacs.admin.ch/en/25-mp2-en (accessed September 2026). ↩
4. National Cyber Security Centre, “Legal basis for the reporting obligation”, https://www.bacs.admin.ch/en/legal-basis-for-the-reporting-obligation (accessed September 2026). ↩
5. Célian Hirsch, “Die neue Meldepflicht wird konkreter”, cdbf.ch, 27 May 2024, https://cdbf.ch/de/1351/ (“Zu diesen Infrastrukturen gehören insbesondere Banken, Versicherungen, Finanzmarktinfrastrukturen oder auch Cloud-Anbieter (Art. 74b ISG).”); National Cyber Security Centre, “Legal basis for the reporting obligation” (above), on energy suppliers, transport companies and cantonal and communal administrations. ↩

For general information only. This article does not constitute legal or tax advice, does not replace advice on an individual case and creates no lawyer-client relationship. It reflects the legal position as of September 2025.