Data protection, outsourcing and cloud computing

The revised Federal Act on Data Protection

The revised Federal Act on Data Protection (FADP) entered into force on 1 September 2023. It is technology-neutral and therefore applies directly to AI-supported processing of personal data. The Federal Data Protection and Information Commissioner (FDPIC) confirmed this in November 2023 and again in an updated statement of May 2025. Manufacturers, providers and deployers of AI systems must make the purpose, functioning and data sources of their processing transparent. Data subjects have the right to object to automated processing and, under Article 21 FADP, to request that an automated individual decision be reviewed by a natural person. They must also be told whether they are interacting with a machine and whether their data are used to train self-learning systems. Where AI applications involve high risks, a data protection impact assessment under Article 22 FADP is required. The FDPIC added that applications designed to undermine privacy, such as comprehensive real-time facial recognition or social scoring, are prohibited under Swiss law.¹

The FDPIC has since applied these principles in individual cases. On Data Protection Day 2026 it reported that it had examined the use of posts on the platform X for training the AI system Grok, with the result that users could object to such use. It also concluded an investigation into intelligent video surveillance at the checkouts of a Swiss retailer. The Commissioner summarised the position by stating that generative AI as such is not subject to data protection law, but becomes so as soon as it processes personal data.²

The sanctions regime of the FADP differs from that of the European General Data Protection Regulation. It provides for criminal fines of up to CHF 250,000 against natural persons who wilfully breach certain duties, among them duties of information and access, duties of care in the engagement of processors and the rules on cross-border disclosure. Fines are not imposed on companies. Several offences are prosecuted only on complaint, which must be filed within three months.³ For compliance officers and managers in Zurich institutions who decide on the introduction of AI tools, the personal nature of this liability is an incentive to document their decisions with care.

Outsourcing under FINMA Circular 2018/3

When a bank or insurer uses an external AI model or runs AI workloads on a cloud platform, the FINMA rules on outsourcing may apply. FINMA Circular 2018/3 “Outsourcing – banks and insurers” was published in December 2017 and entered into force on 1 April 2018. It replaced rules dating from 1999 and was designed to be principles-based and technology-neutral. The circular requires institutions to take appropriate account of the higher risks of outsourcing abroad. It also requires them to ensure that the institution can be restructured or resolved in Switzerland, which in practice means that essential data must remain accessible from Switzerland at all times.⁴

The circular applies when a service provider independently and on an ongoing basis performs a significant function of the institution. Data storage, the operation of databases and the operation of IT systems can all qualify, which covers most cloud-based AI services. The contract must grant the institution, its audit firm and FINMA unrestricted rights to inspect and audit all relevant information at any time. Where client data are transferred to a country without adequate data protection, safeguards such as recognised contractual clauses are required.⁵ In the context of AI, the inventory required by Guidance 08/2024 and the outsourcing inventory overlap. Many institutions have combined them, so that each external model is assessed both as an AI system and as a potential outsourcing arrangement.

Banking secrecy and the cloud

The most specifically Swiss constraint is banking secrecy. Article 47 of the Banking Act makes it a criminal offence for bank employees and agents to disclose information entrusted to them in their professional capacity. Since the Financial Institutions Act came into force in 2020, a comparable professional secrecy duty applies to other financial institutions such as portfolio managers and fund management companies under Article 69 of that Act.⁶ A cloud provider or AI provider that has access to unencrypted client identifying data therefore raises a question of criminal law. The Swiss Bankers Association has addressed this in its Cloud Guidelines, whose third edition was published in November 2025. The Guidelines treat the storage of data abroad as compatible with banking secrecy if the institution adopts appropriate technical, organisational and contractual measures, among them strong encryption and a contractual duty of the provider to notify the bank of requests from foreign authorities. The third edition takes account of FINMA Circular 2023/1 with its new category of critical data, the revised FADP and the Information Security Act.⁷

A further constraint arises from Article 271 of the Swiss Criminal Code, which prohibits the performance of acts on behalf of a foreign state on Swiss territory without authorisation. A cloud provider that hands over data stored for a Swiss bank to a foreign authority outside the channels of international assistance may expose itself and, depending on the circumstances, the bank to criminal liability.⁸ The American CLOUD Act of 2018, which allows US authorities to demand data held by American providers regardless of where it is stored, is the main scenario discussed in this context. The investments of Microsoft, Google and AWS in Swiss data centres reduce the problem of data location, although they do not eliminate it, because the parent companies remain subject to American law. For AI applications that process client data, this legal risk is one reason why Zurich institutions show interest in open models such as Apertus that can be run on premises or with Swiss providers.

Notes

1. FDPIC, “Update – Current data protection legislation is directly applicable to AI”, 8 May 2025, https://www.edoeb.admin.ch/en/update-current-legislation-directly-applicable-ai. ↩
2. FDPIC, “Data Protection Day 2026: Uses of generative AI and data protection issues”, 28 January 2026, https://www.edoeb.admin.ch/en/data-protection-day-2026. ↩
3. Pestalozzi, “The revised Federal Data Protection Act – sanctions increase as of 1 September 2023”, https://pestalozzilaw.com/en/insights/news/legal-insights/the-revised-federal-data-protection-act-sanctions-increase-as-of-1-september-2023/ (accessed September 2026). ↩
4. FINMA, “FINMA veröffentlicht Outsourcing-Rundschreiben”, 5 December 2017, https://www.finma.ch/de/news/2017/12/20171205-mm-rs-outsourcing/. ↩
5. Baker McKenzie, “Cloud Compliance Center – Switzerland”, https://resourcehub.bakermckenzie.com/en/resources/cloud-compliance-center/emea/switzerland (accessed September 2026). ↩
6. Federal Act on Banks and Savings Banks (Banking Act, SR 952.0), Art. 47, bilingual edition, https://www.droit-bilingue.ch/de-fr/9/95/952.0-47-100.html; Federal Act on Financial Institutions (FinIA, SR 954.1), Art. 69 (“Verletzung des Berufsgeheimnisses”), https://www.droit-bilingue.ch/de-fr/9/95/954.1-69-72.html. Both provisions are also cited in the SBA Cloud Guidelines discussed in the following footnote. ↩
7. Lenz & Staehelin, “Third edition of the SBA Cloud Guidelines”, November 2025, https://www.lenzstaehelin.com/news-and-insights/browse-thought-leadership-insights/insights-detail/third-edition-of-the-sba-cloud-guidelines/. ↩
8. Baker McKenzie, “Cloud Compliance Center – Switzerland”, section on banking secrecy and Article 271 SCC, https://resourcehub.bakermckenzie.com/en/resources/cloud-compliance-center/emea/switzerland (accessed September 2026). ↩

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.