AI and Recruitment in Switzerland: What the FADP Allows in 2026

Yes, you can use artificial intelligence to recruit in Switzerland. Sorting applications, matching profiles to positions, pre-qualifying candidates via chatbot: all of this is lawful, provided you inform candidates, keep the final decision human, and regularly audit the system for bias. The technology has been working for a while now. It's the framework that separates a genuine time saving from a legal risk you alone will bear.

Article published in March 2026, revised on 25 May 2026, reviewed on 7 July 2026.

What AI is already doing in recruitment

By 2026, three uses of AI in recruitment have become established in Swiss companies: assisted application sorting, semantic matching between profiles and positions, and pre-qualification via conversational agents.

Assisted sorting addresses a concrete problem. An attractive position can receive several hundred applications within a few weeks. Today's systems, powered by language models, extract the structured elements of a CV — skills, experience, education — regardless of layout, and then produce a relevance ranking against the target profile. The tool's status is that of an assistant: it proposes an order, the recruiter decides. The time saving is real; I am wary, however, of vendors' quantified promises, because without measuring your own flow, no one can tell you what you will actually gain.

Semantic matching goes a step further than keyword sorting. The model recognises that two different phrasings describe the same skill, and this capability works both ways: identifying the best profiles in your internal talent pool for a new position, or offering a candidate opportunities beyond the one they originally targeted.

Pre-qualification chatbots conduct a structured preliminary interview, around the clock, in several languages — which matters in a country where recruitment happens in French, German and Italian. Their place is strictly upstream: they filter and route, they never decide.

Before buying anything, map your actual flow: volumes, stages, bottlenecks. This is the same groundwork as for any business automation — the tool comes after the flow, never before it.

What does the FADP say when an algorithm sorts applications?

The Federal Act on Data Protection, in its revised version that entered into force on 1 September 2023, precisely governs the use of AI in a recruitment process^[1]. Five obligations structure the practice, and none of them is decorative.

Information comes first: candidates must know, before applying, that automated tools are involved in processing their file — in your website's privacy policy or in the application form, in accessible terms. Then the right of access: a candidate can request the data collected about them and, in certain cases, the algorithmic assessments produced. A system that cannot respond within the legal deadlines is a direct operational risk.

Add to this retention — keeping application files indefinitely, without a legal basis, is not admissible, and your archiving practices must be documented and justified — the data protection impact assessment, required before deploying high-risk automated profiling, and transparency on automated individual decisions, for which the FADP requires informing candidates about the criteria used.

None of this bans AI. All of it bans AI deployed carelessly.

If the system discriminates, who is liable? You are.

Swiss law prohibits unlawful discrimination in hiring, and this prohibition applies fully to algorithm-assisted decisions. Technological mediation creates no exception. "The algorithm decided" is not a valid defence: the employer remains legally responsible for the decisions produced by the systems it deploys, even when the discrimination was not intentional.

This full liability is, in my view, the single most important fact in this whole topic. It turns the choice of software into a business decision: what you are prepared to answer for, if it comes to that, before a court.

Algorithmic bias does not correct itself

Machine learning models reproduce — and can amplify — the biases in their training data. Three manifestations are regularly observed: gender bias, in sectors where historical hiring was unbalanced; age bias, which disadvantages certain age groups on grounds that are correlated but not causal; and language bias, which penalises candidates whose native language differs from that of the job posting, regardless of their actual competence.

None of these drifts is fixed by a model update. The correction is operational: diversify the training data, audit results by demographic segment, maintain effective human oversight over individual decisions, and make criteria transparent so that appeals remain possible. This discipline is built into serious use; it isn't bolted on afterward.

Does the European AI Act concern me too?

The European Regulation on artificial intelligence classifies recruitment AI systems as high-risk systems^[2]. For companies within scope, this means comprehensive technical documentation, a conformity assessment before deployment, effective human oversight, transparency obligations, and continuous monitoring after go-live.

A Swiss company falls within scope as soon as it recruits candidates residing in the European Union, or uses a system marketed by a European vendor. These obligations then stack on top of the FADP framework. The resulting regulatory environment is dense, but predictable: those who build compliance into the initial scoping deliver deployable systems; those who address it at the end of the project pay a substantial catch-up cost.

Five safeguards before deploying

From all of this I draw five safeguards. They fit on one page and separate defensible practices from dangerous automation.

• The final hiring decision stays human. AI pre-selects, ranks, flags; it does not decide.
• Auditing results happens on a schedule. By demographic segment, quarterly at minimum, monthly at high volume.
• Candidate information is explicit. Privacy policy and application form clearly disclose automated processing, without jargon.
• The appeal process is documented. A candidate rejected by an automated system must know how to challenge the decision, and you must be able to respond within a reasonable time.
• Training data is qualified. Vendors' generic datasets match neither the Swiss labour market nor your target profiles.

If your project passes these five filters, it deserves to exist. If it fails even one, the problem isn't the tool — it's your process. This is the same trade-off as any automation solution — framework first, software licence second. And to place recruitment among an SME's other AI projects, the overall framework is laid out in What should a Swiss SME do about AI in 2026?.

A word of honesty about what comes next. I don't recommend any tool by name: this market moves too fast for a recommendation to hold for six months. Nor do I publish a price range, because it would depend on your volume, your sector and your infrastructure. MCVA Cahier No. 4, planned for October 2026, will cover AI across the full range of HR functions in Swiss companies.

Key takeaways

— AI in recruitment is lawful in Switzerland: the FADP requires informing candidates, the right of access, limited retention, and an impact assessment for high-risk profiling. — Liability for a discriminatory decision rests entirely with the employer, algorithm or not. — AI pre-selects and ranks; the hiring decision stays human, and that boundary protects you legally.

FAQ

Can I let a system automatically reject applications? Technically yes, legally it's the most exposed configuration. An automated individual decision triggers specific information obligations under the FADP, and you remain liable for every discriminatory rejection. Keeping human judgment over every negative decision is the most defensible position — and the healthiest one for recruitment quality.

Do I have to tell candidates that I use AI? Yes, upfront and in accessible terms. Your recruitment website's privacy policy and the application form must disclose automated processing. A mention buried in unreadable terms and conditions does not satisfy the transparency requirement.

I hire five people a year. Does this concern me? At this volume, the gain from a sorting tool is marginal — but the framework still applies, because your recruitment platform may already embed AI features you never chose. Check what your current tools actually do with applications. Compliance doesn't depend on your size.

Does the AI Act apply if I have no European clients? The regulation looks at who you recruit and where your tool comes from, more than at your client base. As soon as you recruit candidates residing in the EU, or your system is marketed by a vendor on the European market, its obligations kick in. For strictly Swiss recruitment with a tool outside that scope, the FADP remains your primary framework.

And in your own recruitment? The AI Usage Diagnostic: sixty minutes to map your actual workflows, identify what deserves custom development, what stays SaaS, and what needs no AI at all. Book a diagnostic

Sources

[1] Federal Act on Data Protection (FADP), revision of 25 September 2020, in force since 1 September 2023. www.fedlex.admin.ch/eli/cc/2022/491/fr [↩]

[2] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (EU AI Act). eur-lex.europa.eu/eli/reg/2024/1689/oj/eng [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.