Emineon

Trust & governance

Swiss firms that use AI answer to the nFADP, to sector supervisors such as FINMA and, when they serve the EU market, to the EU AI Act. This page summarises what each asks of an AI system in practice, and the controls we build into each system so the evidence exists from day one.

FINMA
FINMA Guidance 08/2024
nFADP
nFADP (revised Federal Act on Data Protection)
EU
EU AI Act
CH
Switzerland’s own approach

01 · The frameworks

A practitioner’s summary for orientation. It is not legal advice. Check the primary sources and your counsel for the obligations that apply to you.

01

FINMA Guidance 08/2024

Banks, insurers, wealth managers and other FINMA-supervised institutions.

FINMA’s guidance on governance and risk management when supervised institutions use AI. It describes what the supervisor expects to see, and it frames supervisory dialogue on AI.

  • Know which AI applications you run, with a maintained inventory and clear owners.
  • Classify each application by risk and manage it accordingly.
  • Look after data quality, testing and ongoing monitoring.
  • Keep documentation that makes results explainable and reviewable.

Source: FINMA · Guidance 08/2024

02

nFADP (revised Federal Act on Data Protection)

Any organisation processing personal data in Switzerland.

In force since 1 September 2023. It applies to AI systems as it does to any processing of personal data, with a few duties that matter most for AI.

  • Privacy by design and by default.
  • Transparency, including about automated individual decisions.
  • A record of processing activities (with an exemption for some smaller, low-risk firms).
  • Data protection impact assessments for high-risk processing, and breach notification to the FDPIC.

Source: Swiss Confederation · nFADP · FDPIC guidance

03

EU AI Act

Swiss firms that place AI systems on the EU market, or whose AI output is used in the EU.

A risk-based EU regulation with obligations that phase in over several years. It can reach Swiss companies through their EU business, even though Switzerland is not an EU member.

  • Prohibited practices, high-risk systems and transparency duties are treated differently.
  • High-risk systems need risk management, data governance, documentation, logging and human oversight.
  • Some duties fall on providers, others on deployers. Know which role you play.
  • Application dates have moved before. Check the current timetable for your use case.

Source: European Union · AI Act

04

Switzerland’s own approach

Everyone operating in Switzerland.

Switzerland has chosen not to copy the EU AI Act. The Federal Council’s approach is to ratify the Council of Europe Convention on AI and to adapt Swiss law sector by sector where needed.

  • Existing law (data protection, sector supervision, liability) already applies to AI today.
  • Further amendments are expected through the normal consultation process.
  • Designing for transparency and human oversight now keeps future changes cheap.

Source: Federal Council · OFCOM (BAKOM)

02 · At a glance

FrameworkConcernsAsks for, in shortWhat we build in
FINMA 08/2024Supervised financial institutionsInventory, risk classification, data quality, testing, monitoring, documentationSystem register, risk tiers, eval suites, monitoring and alerting, decision logs
nFADPAnyone processing personal data in SwitzerlandPrivacy by design, transparency, records, impact assessments, breach notificationData maps, minimisation, Swiss or tenant hosting, DPIA inputs, audit trail
EU AI ActSwiss firms with EU market exposureRisk-based duties, such as documentation, logging and human oversight for high-risk usesTechnical documentation, event logging, human review queues, role mapping
Swiss approachEveryone in SwitzerlandExisting law now, sector-specific amendments laterTransparent, reviewable systems that adapt without a rebuild

03 · Controls we build

Governance is cheapest when it is part of the architecture. These controls come with each Emineon system, sized to its risk.

01

AI inventory

Each system, model and prompt version is registered with an owner, a purpose and a risk tier.

02

Evaluations

We write an eval set before the first prompt. It runs on each change and in production.

03

Logging and audit trail

Inputs, outputs, tool calls and approvals are logged. Any result can be traced and explained.

04

Human review

People approve the important steps. Thresholds and queues are set with the teams who own the decision.

05

Data residency

Swiss-hosted or in your own tenant when required. Data is minimised and not used to train third-party models.

06

Documentation

Technical documentation, data maps and DPIA inputs kept current as the system changes.

06 · FAQ

FINMA Guidance 08/2024 sets out the Swiss financial regulator’s supervisory expectations for governance and risk management when supervised institutions use AI. In practice that means knowing which AI applications you run, who is responsible for them, how risky each one is, and how data quality, testing, monitoring and documentation are handled. We build those controls into the system itself.

It can. The EU AI Act reaches providers that place AI systems on the EU market and, in some cases, providers and deployers outside the EU whose AI output is used in the EU. Switzerland is not copying the EU Act. The Federal Council has chosen to ratify the Council of Europe AI Convention and adapt Swiss law sector by sector.

The revised Federal Act on Data Protection has been in force since 1 September 2023. For AI it means privacy by design and by default, transparency about automated decisions, a record of processing activities, impact assessments for high-risk processing and breach notification to the FDPIC. We treat these as design inputs from the start.

Yes. Where required we deploy on Swiss-hosted infrastructure or in your own cloud tenant, and we design for the nFADP from the start. Your data stays yours and is not used to train third-party models.

No. We are model-agnostic and choose the model per task on quality, cost and data constraints, so you can switch providers without rebuilding.

Agentic AI systems use language models to plan and carry out multi-step tasks with tools, such as reading documents, checking systems and drafting actions, with people approving where it matters. It is ready for production when clear scope, evaluations, logging and human review are built in from day one.

This page is a practitioner’s summary for orientation, not legal advice. Regulatory texts and timelines change. Check the primary sources (FINMA, FDPIC, the Federal Council and the EU) and your legal counsel before relying on it.

Next step

Tell us about one process that takes your team a lot of time. Happy to have a 30-minute call, no preparation needed.