Modern office towers in Zurich financial district on a clear morning

AI Companies Compliance Switzerland GDPR

An AI company based in Zurich with users in Berlin has a compliance problem that is easy to state and fiddly to solve: it answers to two data protection regimes at once. Switzerland’s Federal Data Protection Act (nFADP) governs it at home, the EU’s GDPR follows its European users, and if it places AI systems on the EU market, the EU AI Act arrives on top. The good news is that the two data laws overlap far more than they conflict – Switzerland designed its revised act to stay close enough to the GDPR to keep its EU adequacy status. The work, then, is not doing everything twice. It is building one compliance system at GDPR level, then closing a short, specific list of Swiss gaps. This guide is that list, in the order a company should actually work through it.

In short: Most AI companies active in Switzerland and the EU must comply with both the nFADP and the GDPR at the same time. The practical route is a single GDPR-grade compliance programme – data map, documented legal bases, DPIAs, processor contracts, breach playbook – plus five Swiss-specific additions: privacy notices that name every country data flows to, information and human-review duties for automated decisions, a Swiss representative if you are a foreign company targeting Swiss users, awareness that intentional breaches create personal criminal liability for named individuals (up to CHF 250,000), and a diary note for spring 2027, when Switzerland’s first AI consultation draft and the EU AI Act’s next phase both land.

First, work out which laws actually apply to you

Compliance starts with scope, and for AI companies the answer is usually “more than one”. If you process the personal data of people in Switzerland, the nFADP applies – it has applied, in its revised form, since September 2023, supervised by the federal regulator, the FDPIC. If you offer goods or services to people in the EU, or monitor their behaviour, the GDPR applies to that processing even if your company has never had an office in Europe; its territorial reach is deliberately long. Selling AI systems into the EU market adds the EU AI Act, whose transparency provisions have applied since August 2026. And sector regulators sit on top of all of it: a Swiss AI company in finance also answers to FINMA, which published dedicated AI governance guidance back in December 2024. If you want the law-by-law comparison behind this checklist, our breakdown of the swiss federal data protection act vs gdpr for ai 2027 covers the two regimes side by side.

The compliance checklist

1. Map your data before anything else

You cannot comply with either law for data you do not know you hold. For an AI company the map has to cover training data (where it came from, whose it is, what rights attach to it), prompts and uploaded files from users, outputs that contain personal data, and the logs nobody remembers until a regulator asks. Include where each dataset physically lives and which countries it travels to – that detail becomes a legal duty in step 3. The known what challenges does generative ai faces with respect to data – provenance, consent, purpose drift – are compliance problems first and engineering problems second.

2. Document legal bases the GDPR way

The GDPR requires a named legal basis for every processing operation; Swiss law is more permissive and starts from “allowed unless it breaches the principles”. Build to the stricter standard: a register that states, per purpose, whether you rely on consent, contract, legal obligation or legitimate interests, with the reasoning written down. A GDPR-grade register satisfies Swiss expectations almost automatically, while the reverse is not true.

3. Write privacy notices that satisfy both laws

The two notice regimes look similar but are not identical. The specifically Swiss traps: your notice must name the countries to which personal data is disclosed – an EU template that says “trusted partners worldwide” fails this – and if an automated decision produces legal or significant effects for a person, Swiss law expects that fact, and the person’s right to human review, to be communicated. Draft one notice with a Swiss annex rather than two drifting documents.

4. Build automated-decision compliance into the product

Both laws protect people from purely machine-made decisions, with different mechanics: the GDPR grants a right not to be subject to such decisions outside narrow exceptions, while the nFADP requires that the person be informed and able to obtain a human review and state their view. For an AI product this is a design requirement, not a policy paragraph: the review route has to exist in the interface, a human has to actually be able to overturn the output, and “high-risk profiling” – assembling data into a full personality picture – needs extra care and transparency under Swiss law. If your roadmap includes credit, hiring, insurance or eligibility decisions, treat this step as the centre of the whole programme.

Professional working through a compliance checklist on paper beside a laptop in an office

5. Run DPIAs early – and use the Swiss shortcut properly

High-risk processing – large-scale profiling, sensitive data, novel AI systems making significant decisions – triggers a data protection impact assessment under both laws. Do it at design stage, when changing the system is still cheap. Switzerland offers one genuine procedural gift: appoint an independent data protection advisor, have them sign off the DPIA, and you can be excused from consulting the FDPIC even where a residual high risk remains. Under the GDPR, that same residual risk means mandatory consultation with the regulator. The shortcut rewards companies that build the advisor relationship before they need it.

6. Appoint the right people, in the right countries

Three appointments cover most cases. A data protection officer where the GDPR requires one (large-scale monitoring or sensitive-data core activities capture many AI firms). A Swiss data protection advisor – voluntary for private companies, but the key that unlocks the DPIA shortcut in step 5. And representatives: a foreign AI company that offers services to people in Switzerland professionally and at scale must appoint a representative in Switzerland whom the FDPIC and data subjects can address; the GDPR has the mirror-image rule for non-EU companies targeting Europeans. Missing representative appointments are one of the most common, and most easily checked, compliance gaps.

7. Prepare two breach clocks

The GDPR gives you 72 hours from awareness to notify the supervisory authority. The nFADP has no fixed deadline – notify the FDPIC “as soon as possible”, and only where the breach likely creates a high risk for those affected. One incident can start both clocks at once, on different terms. The playbook that survives contact with reality states plainly: who declares an incident, who drafts for which regulator, and what evidence is preserved in the first hours – because “as soon as possible” is judged in hindsight.

8. Get contracts and transfers in order

Processor agreements with your cloud, model and analytics providers are mandatory homework under both regimes. Transfers between the EU and Switzerland are the easy part – the EU’s adequacy decision for Switzerland means no extra safeguards are needed on that corridor. Transfers onward to the United States or other non-adequate countries need the usual machinery of standard contractual clauses and transfer assessments. Where your models physically run matters here as well: the argument behind why are data centers controversial is partly a data-location argument, and your transfer paperwork has to match the infrastructure reality.

Engineer inspecting server racks inside a secure Swiss data center

The mistakes AI companies keep making

The same four errors recur. First, copying a GDPR programme and assuming Switzerland is covered – it mostly is, except for the representative, the country-disclosure duty, and the personal-liability sting, which are exactly the parts nobody copied. Second, the mirror error: assuming that because Swiss law is lighter, Switzerland means “no compliance” – the FDPIC has confirmed the FADP applies fully to AI processing, and intentional breaches of information, cooperation or security duties are criminal matters for the responsible individuals, with fines up to CHF 250,000 that no company budget absorbs on their behalf. Third, treating user prompts and training corpora as if they were not personal data at all, when both laws will treat much of them as exactly that. And fourth, ignoring the EU AI Act because the company is “Swiss” – the Act follows the product into the EU market, not the passport of its maker.

What to watch in 2027

Two dates deserve a place in the compliance calendar now. In spring 2027, Swiss federal departments are due to publish the consultation draft implementing the Council of Europe AI Convention – signed by Switzerland in March 2025 – covering transparency, data protection, non-discrimination and supervision; it is the first concrete text of Switzerland’s sector-by-sector AI framework, and companies that comment on drafts tend to fare better than companies that meet final law. In the same period, the EU AI Act’s phase-in continues to its high-risk core, and Switzerland hosts a global AI summit in Geneva to promote its middle-way model. None of this rewrites the checklist above – but all of it will add footnotes to it, and the companies that built GDPR-grade systems with the Swiss deltas closed will absorb those footnotes in an afternoon.

Frequently asked questions

Does an AI company in Switzerland need to comply with the GDPR?

Only if it processes data of people in the EU in the ways the GDPR reaches for – offering them goods or services, or monitoring their behaviour. A purely Swiss-facing company answers to the nFADP alone. In practice, most AI companies with any European user base end up complying with both, which is why building one GDPR-grade system plus the Swiss additions is the standard approach.

What is the first compliance step for an AI company under both laws?

Data mapping. Neither regime can be satisfied for processing you cannot describe: what personal data you hold (including training data, prompts and logs), why you hold it, where it lives, which countries it reaches and who else touches it. Every later artefact – legal-basis register, privacy notices, DPIAs, transfer paperwork – is built on that map.

Does a foreign AI company need a representative in Switzerland?

In many cases, yes. A foreign controller that offers goods or services to people in Switzerland on a professional, larger scale must appoint a representative in Switzerland who serves as the contact point for the FDPIC and for data subjects. It is one of the most commonly missed Swiss requirements – and one of the easiest for the regulator to verify.

Are DPIAs mandatory for AI systems in Switzerland?

A data protection impact assessment is required under the nFADP when planned processing is likely to create a high risk for people’s personality or fundamental rights – a description that captures many profiling and automated-decision systems. The GDPR has the equivalent duty. Switzerland’s distinctive feature is procedural: an independent data protection advisor’s sign-off on the DPIA can excuse the company from consulting the FDPIC about residual risks.

If we fully comply with the GDPR, are we automatically compliant in Switzerland?

Mostly, but not automatically. GDPR compliance covers the great majority of Swiss expectations because the nFADP was aligned to it. The recurring gaps are Swiss-specific: naming the destination countries in privacy notices, the information and human-review duties around automated decisions in their Swiss form, appointing a Swiss representative where required, and remembering that intentional breaches expose named individuals to personal criminal fines rather than the company to an administrative one.

Will Switzerland add AI-specific compliance duties in 2027?

New duties are coming, but gradually and sector by sector, not as a Swiss copy of the EU AI Act. Switzerland signed the Council of Europe AI Convention in March 2025 and plans to implement it mainly for public-sector actors, with a consultation draft expected in spring 2027 covering transparency, data protection, non-discrimination and supervision. Private AI companies should expect targeted additions to existing law – the FADP included – rather than a single new AI statute.

Leave a Reply

Your email address will not be published. Required fields are marked *