Swiss and European Union flags outside a modern office building in Zurich

Swiss Federal Data Protection Act vs GDPR for AI 2027

If your AI company sells into Europe, Switzerland looks deceptively simple. It is not in the EU, its data protection law was deliberately modelled close to the GDPR, and personal data flows between the EU and Switzerland without extra paperwork thanks to an EU adequacy decision. So far, so comfortable. Then you read the details, and the two systems turn out to disagree about some fundamental things: who gets punished, whether you need permission to process data at all, and what happens when an algorithm makes a decision about a person. With the EU AI Act phasing in on one side of the border and Switzerland preparing its first AI consultation draft for spring 2027 on the other, 2027 is the year these differences stop being academic. This is the comparison AI companies actually need.

In short: The GDPR starts from prohibition – processing personal data is unlawful unless a legal basis allows it – while the Swiss Federal Data Protection Act (nFADP, in force since September 2023) starts from permission: processing is generally allowed unless it breaches the law’s principles or a person’s rights. The GDPR fines companies up to €20 million or 4% of global turnover; Switzerland criminally fines the responsible individuals up to CHF 250,000, and only for intentional violations. Both laws demand impact assessments for high-risk AI and give people rights over automated decisions, but the mechanics differ. And while the EU layers the AI Act on top of the GDPR, Switzerland has no AI law – it signed the Council of Europe AI Convention in March 2025, chose a sector-by-sector approach, and expects a consultation draft in spring 2027.

Two laws, two opposite starting points

The deepest difference is philosophical, and it shapes everything else. Under the GDPR, every act of processing personal data needs a legal basis from a closed list – consent, contract, legal obligation, legitimate interests and so on. No basis, no processing. For AI companies, that single rule drives the hardest questions in the industry: what is your legal basis for training on this data, for profiling these users, for feeding prompts into a model?

The nFADP flips the default. Swiss law treats processing by private companies as generally permissible. You do not need to point to a permission slip for each operation; you need to respect the principles – lawfulness, good faith, proportionality, purpose limitation, accuracy, security – and you must stop or justify yourself when someone objects or when sensitive data and profiling raise the stakes. In practice, a Swiss AI product can do things by default that would need a carefully documented legitimate-interests argument in the EU.

The bridge between the two worlds is adequacy: the EU recognises Swiss protection as essentially equivalent, so EU–Swiss data transfers need no extra safeguards. That is precisely why Switzerland keeps its law GDPR-shaped – drifting too far would endanger the adequacy finding its export economy depends on.

The differences that actually bite AI companies

Training data and everyday processing

In the EU, an AI company must be able to name its Article 6 legal basis for training data and defend it – regulators and the European Data Protection Board have spent years arguing over whether “legitimate interests” can carry large-scale model training at all. In Switzerland, the same training run starts from “allowed unless it breaks the principles”, with transparency duties doing most of the work. It is a lighter lift, though not a free pass: the Swiss regulator, the FDPIC, has confirmed the FADP applies fully to AI-supported processing, and the wider what challenges does generative ai faces with respect to data – consent, secrecy about sources, purpose creep – are exactly where companies get into trouble first.

Automated decisions and profiling

Here both laws protect the individual, but with different mechanics. GDPR Article 22 starts as a near-prohibition: a person has the right not to be subject to a purely automated decision with legal or similarly significant effect, subject to exceptions wrapped in safeguards. The nFADP instead imposes duties around the decision: if an automated decision has legal effect or significantly affects someone, you must inform them, and on request a human must review it and the person gets to state their case. Switzerland also has a concept the GDPR lacks – “high-risk profiling”, profiling that assembles data into a full picture of someone’s personality – which attracts stricter consent and transparency treatment. If your AI scores, ranks or classifies people, this is the Swiss provision to lose sleep over. Building systems whose decisions can be explained to the human reviewer is where explainable AI stops being a research topic and becomes a compliance feature.

Impact assessments

Both regimes require a data protection impact assessment (DPIA) when processing is likely to be high risk – new AI systems making significant decisions are the textbook trigger in both. One Swiss twist is genuinely company-friendly: if you appoint an independent data protection advisor and the advisor signs off on your DPIA, you can be excused from consulting the FDPIC even when a residual high risk remains. Under the GDPR, residual high risk after a DPIA means mandatory prior consultation with the regulator. Same assessment, very different ending.

The breach clock

The GDPR’s 72-hour deadline to notify the supervisory authority after a personal data breach is famous, and feared. The nFADP has no fixed clock: notify the FDPIC “as soon as possible”, and only when the breach is likely to create a high risk for the people affected. Less panic, but also less certainty – “as soon as possible” is an argument waiting to happen after the fact.

Officers and representatives

A DPO is mandatory under the GDPR in defined cases, which capture many AI companies (large-scale monitoring or sensitive data as a core activity). In Switzerland, appointing a data protection advisor is voluntary for private companies – though, as the DPIA route shows, doing so buys real procedural advantages. Foreign AI companies targeting Swiss users at scale may instead need the other Swiss appointment: a local representative in Switzerland, the FDPIC’s address for complaints. Website operators should also note the data-flow symmetry here: Swiss notices must name the countries data travels to, and users on both sides of the border increasingly want to know exactly where their data goes.

Business team in a modern office reviewing compliance documents and charts on a screen

Who pays when it goes wrong

This is the difference executives feel in their stomachs. GDPR enforcement aims at the company: administrative fines up to €20 million or 4% of worldwide annual turnover, whichever is higher, imposed by regulators who can also ban processing outright. The bill lands on the balance sheet.

Swiss enforcement aims at the human. The FDPIC cannot fine your company. Instead, the nFADP makes intentional breaches of specific duties – information duties, cooperation with the regulator, minimum data security – a criminal matter for the responsible individuals: directors, managers, the person who actually decided. The maximum is CHF 250,000, far below a GDPR fine, but it is personal, it is criminal, and it follows the individual rather than dissolving into a corporate budget. Swiss compliance conversations are therefore strangely personal: the question is never just “what will this cost the company?” but “whose name is on this decision?”

The AI Act-shaped hole, and what fills it in 2027

On the EU side, data protection is only the first layer. The EU AI Act sits on top of the GDPR, and its obligations are phasing in on a fixed calendar – its transparency provisions have applied since August 2026, with the high-risk machinery following – the same transparency instinct behind the how to remove ai info from instagram story labels users now see on social platforms. Any Swiss company placing AI systems on the EU market is caught by it regardless of what Bern does.

Switzerland, meanwhile, has deliberately chosen not to write an AI Act. Its strategy, set by the Federal Council in February 2025, is to ratify the Council of Europe’s Framework Convention on AI – signed by Switzerland in March 2025 – implement it mainly for state actors, and otherwise legislate sector by sector, with cross-sector rules only where fundamental rights demand them. Finance already lives this model through FINMA’s AI governance guidance. The next milestone is legislative: the federal departments are preparing a consultation draft on the AI rules for spring 2027, covering transparency, data protection, non-discrimination and supervision – the same spring Switzerland hosts a global AI summit in Geneva to present its middle-way approach. In other words, 2027 is when Switzerland’s AI framework starts becoming text that companies can plan against, while the EU’s framework finishes becoming enforceable reality.

Geneva lakefront and international district under a clear sky in Switzerland

What an AI company selling into both markets should do

The practical strategy most dual-market companies land on is unglamorous: build to the GDPR as your ceiling, then bolt on the Swiss deltas. Concretely, that means documenting a legal basis for training and profiling even though Switzerland does not demand one in the same form; designing automated-decision flows that inform people and route to a human reviewer, which satisfies both Article 22’s safeguards and the nFADP’s information-and-review duties; running DPIAs early, and if you operate seriously in Switzerland, appointing the voluntary Swiss advisor to unlock the lighter FDPIC route. Keep your Swiss privacy notice honest about the countries data flows to – a specifically Swiss transparency point that EU templates often miss – and decide, in writing, who the responsible individuals are, because under Swiss law those names are where liability lives. Finally, diarise spring 2027: the Swiss consultation draft and the AI Act’s continuing phase-in will both move the goalposts in the same season, and the companies that read drafts early rarely regret it.

Frequently asked questions

Does the GDPR apply in Switzerland?

Not directly. Switzerland is not an EU or EEA member, so the Swiss Federal Data Protection Act governs processing in Switzerland. But the GDPR still reaches Swiss companies that offer goods or services to people in the EU or monitor their behaviour – and because the EU grants Switzerland an adequacy decision, data can flow between the two without extra transfer safeguards.

Which law is stricter for AI companies – the Swiss FADP or the GDPR?

Overall, the GDPR. It requires a legal basis for every processing operation, imposes fines of up to 4% of global turnover on companies, and is now layered with the EU AI Act. The FADP is lighter on process but has one sharper edge: intentional violations can bring personal criminal fines of up to CHF 250,000 for the responsible individuals.

What are the fines under the Swiss FADP compared with the GDPR?

Under the GDPR, regulators fine the company up to €20 million or 4% of worldwide annual turnover, whichever is higher. Under the Swiss FADP, there are no corporate administrative fines; instead, individuals responsible for intentional breaches of duties such as information, cooperation or data security face criminal fines of up to CHF 250,000.

Does Switzerland have an AI law like the EU AI Act?

No. Switzerland has no overarching AI legislation and has chosen a sector-specific approach instead: it signed the Council of Europe AI Convention in March 2025, plans to implement it mainly for public-sector actors, and relies on existing law – the FADP, sector rules such as FINMA’s AI guidance – for private companies. A consultation draft on AI rules is expected in spring 2027. Swiss companies selling AI into the EU must still comply with the EU AI Act itself.

Do AI companies need a data protection officer in Switzerland?

For private companies, appointing a data protection advisor is voluntary under the FADP – unlike the GDPR, where a DPO is mandatory in defined cases that capture many AI firms. Voluntary appointment still pays off in Switzerland: an independent advisor’s sign-off on a DPIA can excuse the company from consulting the FDPIC about residual high risks.

How do automated decision rules differ for AI under the two laws?

The GDPR gives people a right not to be subject to purely automated decisions with legal or similarly significant effect, with narrow exceptions and safeguards. The Swiss FADP takes a duties-based route: the company must inform the person about such a decision, and on request a human must review it and hear the person’s point of view. Switzerland additionally singles out “high-risk profiling” – building a full personality picture from data – for stricter treatment.

Leave a Reply

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