Contents
    Risk pyramid of the EU AI Act with four tiers: Unacceptable, High, Limited, Minimal
    The four risk classes of the EU AI Act.

    The EU AI Act does not automatically apply to every Swiss SME, but it reaches more Swiss companies than many expect. What matters is not where you are based but whether an AI system or its output ends up in the EU. Since 2 August 2026 the transparency obligations for chatbots, deepfakes and certain AI-generated texts have applied; the Digital Omnibus pushed the high-risk rules back to December 2027 and August 2028. This guide explains, from a Swiss perspective, when you are affected, what applies today, what comes next and what Swiss data protection law already requires.

    Does the EU AI Act apply to Swiss companies?

    Switzerland is not an EU member, and the AI Act is not Swiss law. But the Act is tied to the market, not to a company's seat. Art. 2(1) covers, among others:

    • providers placing AI systems on the market or putting them into service in the EU, "irrespective of whether those providers are established or located within the Union or in a third country" (point (a));
    • deployers established or located in the EU (point (b));
    • providers and deployers in third countries where the output of their AI system is used in the EU (point (c));
    • importers and distributors of AI systems (point (d));
    • affected persons located in the EU (point (g)).

    For Swiss companies, point (c) is the key sentence:

    "providers and deployers of AI systems that have their place of establishment or are located in a third country, where the output produced by the AI system is used in the Union"

    Art. 2(1)(c) AI Act

    That gives four typical triggers for a Swiss SME: you place an AI system on the EU market; the output of your AI system is used in the EU; you have a subsidiary or branch in the EU that uses AI; or customers in the EU interact directly with your AI. Purely private use, by the way, does not make anyone a deployer; the definition excludes personal, non-professional activity.

    Typical situations for Swiss SMEs

    The following is our reading of the regulation, not an official interpretation:

    SituationRoleAssessment
    Zurich software company sells an AI tool to customers in GermanyProviderin scope (Art. 2(1)(a))
    Fiduciary firm uses ChatGPT only internally, clients only in SwitzerlandDeployer with no EU linkoutside the AI Act, the FADP applies
    Web shop with a chatbot that also serves consumers in the EUprovider or deployer, depending on the set-upArt. 50(1) transparency duty relevant
    Swiss group whose German subsidiary runs a customer chatbotthe subsidiary is an EU deployerin scope (Art. 2(1)(b))
    Recruitment agency uses AI to pre-screen applications for client companies in the EUdeployer of a high-risk systemoutput used in the EU (point (c)), high-risk obligations from 2 December 2027
    Agency delivers AI-generated content to clients in the EUDeployergrey area, open question of interpretation

    The grey area: content for clients in the EU

    Many Swiss service firms use AI for clients in Germany or Austria: the agency writing blog posts, the fiduciary preparing analyses. On the wording alone, AI output is used in the EU here. Whether Art. 2(1)(c) really covers every such case has not been settled, and we are not aware of any binding interpretation.

    Our pragmatic approach: where it costs little, we act as if Art. 50 applied. Deepfakes get labelled, texts are reviewed by a human, and who holds editorial responsibility is documented. That is good practice anyway.

    Two more points. Providers of high-risk AI from third countries need an authorised representative in the EU (Art. 22), as do providers of general-purpose AI models (Art. 54); for other AI systems this is not required. And how EU authorities would enforce obligations against Swiss companies with no EU presence is an open question. Relying on that is not a strategy.

    Provider or deployer: which role does your SME have?

    The AI Act assigns obligations by role. The two that matter most:

    • Providers (Art. 3(3)) develop an AI system, or have one developed, and place it on the market or put it into service under their own name or trademark, whether for payment or free of charge.
    • Deployers (Art. 3(4)) use an AI system under their own authority.

    The definition of a deployer, word for word:

    "‘deployer’ means a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity"

    Art. 3(4) AI Act

    There are also importers (Art. 3(6)) and distributors (Art. 3(7)). Most Swiss SMEs are deployers: they use ChatGPT, Claude, Copilot or AI features in their software. You become a provider as soon as you offer your own AI application under your name, such as a self-built advice bot.

    When does a deployer become a provider?

    This is the biggest trap for SMEs. Under Art. 25(1), a deployer is considered the provider of a high-risk AI system if it

    1. puts its name or trademark on a high-risk AI system,
    2. makes a substantial modification to a high-risk AI system so that it remains high-risk, or
    3. modifies the intended purpose of a system so that it becomes high-risk.

    The third case, word for word:

    "they modify the intended purpose of an AI system, including a general-purpose AI system, which has not been classified as high-risk and has already been placed on the market or put into service in such a way that the AI system concerned becomes a high-risk AI system in accordance with Article 6."

    Art. 25(1)(c) AI Act

    The classic example: an SME takes a general-purpose chatbot and has it systematically assess and rank job applications. A general tool becomes a system for selecting applicants, which is a high-risk use case under Annex III. The SME then takes on the provider obligations under Art. 16, including conformity assessment. Those obligations only bite from 2 December 2027, but the decision about how you use AI in recruiting is made today.

    The Digital Omnibus made Art. 25(2) more specific: the original provider must in particular give the new provider technical documentation, information on known limitations and failure modes, and targeted technical access, unless it has clearly ruled out conversion into a high-risk system. Breaches of these duties are now expressly subject to fines (Art. 99(4)(da)).

    What applies when? The timeline

    DateWhat appliesRelevance for Swiss SMEs
    1 August 2024AI Act enters into forcestarting point
    2 February 2025Prohibited practices (Art. 5) and AI literacy (Art. 4)applies today
    2 August 2025Rules for general-purpose AI models, governance, penalty rulesmainly affects model providers
    27 July 2026Digital Omnibus (Reg. 2026/1744) in forcenew dates and relief
    2 August 2026General application, transparency obligations (Art. 50), national enforcementapplies today
    2 December 2026Two new prohibitions; deadline for machine-readable marking (Art. 50(2)) for generative AI placed on the market before 2 August 2026affects providers of generative AI
    2 August 2027National AI regulatory sandboxes; deadline for general-purpose AI models placed on the market before 2 August 2025little direct relevance
    2 December 2027High-risk AI under Annex III (Chapter III, Sections 1 to 3)recruiting, performance monitoring, credit scoring, insurance pricing
    2 August 2028High-risk AI under Annex I (regulated products)makers of products with AI safety components
    2 August 2030Legacy high-risk systems used by public authoritiessuppliers to the public sector

    For SMEs within scope this means: Art. 4, Art. 5 and Art. 50 apply today. The deployer obligations for high-risk AI under Art. 26 do not apply yet.

    What did the Digital Omnibus change?

    The Digital Omnibus is Regulation (EU) 2026/1744 of 8 July 2026, published on 24 July 2026 and in force since 27 July 2026. The Commission proposed it on 19 November 2025, and political agreement was reached on 7 May 2026. The key changes:

    1. Fixed dates for high-risk AI. Annex III applies from 2 December 2027, Annex I from 2 August 2028. According to the amending regulation, the reason is that standards and national authorities were not ready in time (recital 40). Originally the Annex III obligations would have applied from 2 August 2026 and the Annex I obligations from 2 August 2027.
    2. Softer AI literacy. Art. 4 was replaced: instead of ensuring a sufficient level of AI literacy, companies must take measures to support it (details below).
    3. Art. 50 almost unchanged. Only paragraph 7 was amended. A new Art. 111(4) gives providers of generative AI systems placed on the market before 2 August 2026 until 2 December 2026 for machine-readable marking. There is no grace period for chatbot disclosure (para. 1), emotion recognition (para. 3) or deepfakes (para. 4).
    4. More relief for SMEs. SMEs and "small mid-cap enterprises" are now defined (Art. 3(14a) and (14b)). There is simplified technical documentation (Art. 11), a proportionate quality management system (Art. 17(2)), the simplified route for it, now open to all SMEs without partner or linked enterprises (Art. 63(1)), priority access to the AI Office's new EU-level sandbox (Art. 57(3a)) and milder fine rules for small mid-caps (Art. 99(6a)).
    5. Two new prohibitions from 2 December 2026: AI systems that generate realistic intimate imagery of identifiable people without their consent, and AI systems that generate child sexual abuse material.
    6. Smaller points: the fundamental rights impact assessment (Art. 27) can refer to a data protection impact assessment, and a new Art. 4a covers special categories of data for detecting bias.

    The Omnibus did not change the territorial scope in Art. 2(1), the core of the transparency obligations or the SME rule on fines in Art. 99(6).

    What are the risk classes?

    The AI Act classifies AI by risk. The higher the risk to health, safety and fundamental rights, the stricter the obligations:

    Risk classExamplesWhat appliesFrom when
    Unacceptable riskSocial scoring, subliminal manipulation, emotion recognition at work or in education, untargeted scraping of facial imagesprohibited (Art. 5)since 2 February 2025, two more prohibitions from 2 December 2026
    High riskRecruiting and performance monitoring, credit scoring, life and health insurance pricing, emotion recognition (outside work and education), AI in regulated productsstrict obligations for providers and deployersAnnex III from 2 December 2027, Annex I from 2 August 2028
    Transparency riskChatbots, deepfakes, AI texts on matters of public interest, emotion recognitiondisclosure (Art. 50)since 2 August 2026
    Minimal riskSpam filters, simple recommendations, internal draftsno specific obligationsusable now

    Note: supporting AI literacy under Art. 4 applies across all classes, including minimal-risk AI. General-purpose AI models, such as the models behind ChatGPT or Claude, follow a separate track with obligations for the model providers, not for you as a user.

    What does your SME have to comply with today?

    Three areas matter today: AI literacy, the prohibitions and the transparency obligations.

    AI literacy under Art. 4: softer, but not gone

    The original wording required companies to take measures to ensure, to their best extent, a sufficient level of AI literacy. Since the Digital Omnibus, Art. 4(1) reads:

    "Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used. This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual."

    Art. 4(1) AI Act as amended by Regulation (EU) 2026/1744

    In short: you have to do something, but you do not have to guarantee a result. The Commission and the Member States are to support SMEs in particular, and the Commission publishes practical examples on the single information platform (Art. 4(2)). On our reading, Art. 4 is not listed in any of the fine tiers in Art. 99(3) to (5). Member States may, however, lay down penalties, including warnings, for any infringement under Art. 99(1).

    What works as a measure in practice: a short training on the tools your team actually uses, an internal AI policy, a named contact person for questions and a record of who was trained when. We describe how to set this up in our article on AI courses, workshops and programmes, and for public administrations in AI training for municipalities. For SMEs we run our own AI workshops.

    Prohibited practices under Art. 5

    Since 2 February 2025 certain AI practices have been prohibited, including subliminal manipulation, exploiting the vulnerabilities of certain groups, social scoring, emotion recognition in the workplace and in education, and untargeted scraping of facial images from the internet or CCTV footage. The two new prohibitions mentioned above follow from 2 December 2026.

    Most SMEs will never come near these. The one most likely to matter is emotion recognition: a tool that analyses the mood of employees, for example in customer service, needs careful checking before it is introduced. Art. 5 is the most expensive fine tier.

    Transparency obligations under Art. 50

    The transparency obligations have applied since 2 August 2026 and are the most pressing current point for Swiss SMEs with an EU link:

    • Para. 1, chatbots: people must be told that they are interacting with AI unless it is obvious. The duty is addressed to providers.
    • Para. 2, marking: providers of generative AI mark outputs in a machine-readable format. That is normally a job for OpenAI, Google and co., not for an SME that merely uses the images. If you offer your own generative AI system under your own name, such as a self-built text or image generator, you are the provider and para. 2 applies to you.
    • Para. 3, emotion recognition: deployers inform the people concerned.
    • Para. 4, deepfakes and AI texts: deployers disclose deepfakes, and AI texts published to inform the public on matters of public interest.
    • Para. 5, timing: "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure".

    Art. 3(60) defines a deepfake as:

    "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful"

    Art. 3(60) AI Act

    For AI-generated text, the disclosure duty does not apply "where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content" (Art. 50(4), second subparagraph). For an SME that edits its blog posts before publishing, this is the exception that matters.

    Art. 50 in practice: six examples

    UseWho is obligedWhat to do
    Chatbot on your websiteprovider (para. 1); if you build the bot yourself and run it under your own name, on our reading you are often the providera clear notice at the latest with the first message, e.g. "You are chatting with an AI assistant"
    Phone or voice botas for the chatbotnotice at the start of the call
    Realistic AI video of your managing director saying something he never saiddeployer (para. 4)disclose that the content is artificially generated
    AI illustration that clearly does not pretend to show a real sceneno duty under para. 4 if it is not a deepfake (our reading); the tool provider marks it machine-readably (para. 2)nothing extra needed, voluntary labelling possible
    Expert article drafted with AI, reviewed and owned by a personexception under para. 4, second subparagraphdocument the review and responsibility internally
    Automatically generated news on public matters without human reviewdeployer (para. 4)disclose

    On our reading of the wording, ordinary advertising copy and product descriptions are not automatically covered by the text rule, because they do not inform the public on matters of public interest. For borderline cases there are the Commission's Art. 50 guidelines of 20 July 2026 and, since 10 June 2026, a voluntary code of practice on marking and labelling AI-generated content.

    By the way: you need the chatbot notice even without an EU link. Under Swiss data protection law, the FDPIC expects users to know when they are talking to a machine.

    What comes in 2027 and 2028: high-risk AI and deployer duties

    From 2 December 2027, the obligations for high-risk AI under Annex III apply. For SMEs the main areas are recruiting and performance monitoring, credit scoring, and life and health insurance pricing. From 2 August 2028, Annex I follows, covering AI in regulated products.

    Anyone deploying a high-risk system will then have to, under Art. 26 among other things:

    • use the system in line with the provider's instructions for use,
    • assign human oversight to people with the necessary competence, training and authority,
    • ensure that input data under their control is relevant and sufficiently representative for the purpose,
    • monitor operation and report risks and serious incidents,
    • keep automatically generated logs for at least six months,
    • inform workers and their representatives before using the system in the workplace,
    • use the provider's information for the data protection impact assessment.

    Our advice for HR teams and recruiters that use AI to pre-screen applications for roles in the EU: ask your software vendor now whether the tool counts as a high-risk system and what documentation they will provide. Commission guidelines on high-risk classification have been available in draft since May 2026, with the final version planned for the end of 2026.

    What fines can SMEs face?

    Art. 99 sets three tiers:

    BreachMaximumBasic rule
    Prohibited practices (Art. 5)EUR 35 million or 7% of worldwide annual turnoverwhichever is higher (para. 3)
    Certain other obligations (closed list in para. 4), incl. deployers (Art. 26) and transparency (Art. 50)EUR 15 million or 3%whichever is higher (para. 4)
    Incorrect information to authoritiesEUR 7.5 million or 1%whichever is higher (para. 5)

    For SMEs, Art. 99(6) reverses the rule. The lower amount applies:

    "In the case of SMEs, including start-ups, each fine referred to in this Article shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower."

    Art. 99(6) AI Act

    As an illustration: an SME with EUR 2 million annual turnover breaches Art. 50. The candidates are EUR 15 million or 3% of turnover, which is EUR 60,000. For the SME the lower amount applies, so the cap is EUR 60,000. That is a maximum, not a fine that is due automatically.

    The Digital Omnibus left this rule untouched and added a similar one for small mid-caps (Art. 99(6a), for the tiers in paras. 4 and 5). Member States must also take into account the economic viability of SMEs and small mid-caps when imposing penalties, and besides fines, warnings and non-monetary measures are possible (Art. 99(1); the mention of small mid-caps there is new).

    Correction, 14 October 2026: an earlier version of this article wrongly stated that the higher of the two amounts applies to SMEs and overstated the Act's reach for Swiss SMEs. Both have been corrected.

    Does Switzerland have its own AI law?

    No, not as of November 2026. Switzerland is taking its own path:

    • 12 February 2025: the Federal Council decides to ratify the Council of Europe AI Convention and to adapt Swiss law sector by sector. A copy of the EU AI Act is not planned.
    • 27 March 2025: Federal Councillor Albert Rösti signs the Convention in Strasbourg. It has not yet been ratified.
    • 28 October 2026: the Federal Council takes note of the progress of the work. The Federal Department of Justice and Police (Federal Office of Justice), together with the environment, transport, energy and communications department (DETEC) and the foreign affairs department (FDFA), is drafting the binding rules covering, among other things, transparency, data protection, non-discrimination and oversight. The consultation is now planned for spring 2027; it was originally due by the end of 2026.
    • Voluntary measures: the federal government has published a guide that helps associations and companies draw up their own voluntary AI code.

    For SMEs this means: new Swiss AI rules will come at the earliest after the consultation and the parliamentary process. If you use AI today, you follow current Swiss law and, where there is an EU link, the AI Act.

    What already applies in Switzerland? AI and the FADP

    Even without an AI law, AI in Switzerland is not unregulated. According to the Federal Data Protection and Information Commissioner (FDPIC), the revised Federal Act on Data Protection (FADP), in force since 1 September 2023, is technology neutral and directly applicable to AI-supported data processing (first stated on 9 November 2023, updated in 2025). In concrete terms, the FDPIC expects:

    • transparency about the purpose, functioning and data sources of an AI application,
    • telling users when they are talking to a machine and whether their inputs are used for training,
    • labelling deepfakes of identifiable people,
    • a data protection impact assessment for high-risk processing,
    • the right to object to automated decisions and to request human review.

    The FDPIC considers applications that undermine privacy, such as comprehensive real-time facial recognition or social scoring, to be prohibited.

    Two articles are particularly relevant for AI. Art. 21 FADP requires companies to inform people about a decision that is based exclusively on automated processing and that has a legal effect on them or significantly affects them (in the official German text: «die ausschliesslich auf einer automatisierten Bearbeitung beruht und die für sie mit einer Rechtsfolge verbunden ist oder sie erheblich beeinträchtigt»). The person concerned can state their view and request that the decision be reviewed by a human (para. 2); para. 3 sets out exceptions. Art. 22 FADP requires a data protection impact assessment where processing may entail a high risk to the personality or fundamental rights («ein hohes Risiko für die Persönlichkeit oder die Grundrechte»).

    The overlap with the AI Act is large: the FDPIC expects chatbot notices, deepfake labelling and human control even from SMEs with no EU link at all. More in our article on AI and data protection in Switzerland.

    Checklist: what Swiss SMEs should do now

    1. Build an inventory. Which AI tools do we use, including those hidden in CRM, accounting or email software, and for what? Typical use cases are in our article on ChatGPT in the company.
    2. Check the role per use case. Deployer, or, because we build it ourselves or change its purpose, provider? Especially critical: AI in recruiting.
    3. Check the EU link. EU market, output used in the EU, an EU subsidiary, or EU customers interacting directly with our AI?
    4. Implement Art. 50. AI notice in the chatbot, label deepfakes, define human review and editorial responsibility for published texts.
    5. Support AI literacy. Short trainings on the tools in use, a contact person, a record of who was trained when.
    6. Introduce an internal AI policy. Which tools, which data, who approves? How to fit it on one page: internal AI policy.
    7. Review contracts. Data processing, training on your data, data location and, for high-risk uses, the documentation the vendor must provide.
    8. Set a review date. As soon as the Swiss consultation opens, planned for spring 2027, and at the latest before 2 December 2027.

    If you introduce AI agents or process automation, build these points in from the start.

    Where can Swiss SMEs get help?

    The European Commission runs the AI Act Service Desk, a single information platform with the AI Act Explorer for reading the text, a Compliance Checker and a form for questions. This is also where the practical examples on AI literacy under Art. 4 are published. The Commission is also working on further guidance, for example on the value chain, on substantial modification, on simplified quality management for SMEs and on a template for the fundamental rights impact assessment.

    One limitation affects Swiss companies directly: priority access to AI regulatory sandboxes under Art. 62 goes to SMEs "having a registered office or a branch in the Union". A Swiss SME without an EU entity does not get that priority; whether it can take part at all depends on the eligibility criteria of the sandbox concerned.

    On the Swiss side, the FDPIC's guidance on AI and data protection is the first port of call. And if you want to set up AI properly in your business, we help in practice: with AI workshops for SMEs that build AI literacy in your team, and with AI automation that has labelling, human control and documentation built in from the start.

    Conclusion

    The EU AI Act is no reason for Swiss SMEs to panic, but it is not purely an EU matter either. Three things count today: support AI literacy, stay clear of the prohibited practices, and make chatbots, deepfakes and certain AI texts transparent. The high-risk obligations arrive from December 2027; if you use AI in recruiting or credit decisions, prepare now. And much of this is already required by Swiss data protection law anyway.

    Want to know where your SME stands? Talk to us.

    Tenzin Langdun

    About the author

    Co-Founder of Hierarchy · SEO, GEO & AI Expert

    Tenzin Langdun is a co-founder of Hierarchy, where he leads SEO and GEO: how visible companies are in Google, in Google's AI answers and in assistants such as ChatGPT, Gemini and Claude. He also leads the agency's own data studies, including the Zurich AI Visibility Study of 1,515 AI answers about 5,110 companies. He brings over 10 years of marketing experience across strategy, paid acquisition and SEO, and has worked at KPMG, EY, Siemens, Adnovum and LGT, in AI, cybersecurity, audit and consulting, and in software engineering for the insurance sector and for banking and finance. He holds a BSc from ZHAW and an MSc in Artificial Intelligence from the University of Bath. His research on AI-based thyroid cancer detection was published in Scientific Reports (Nature Portfolio) in 2024 and won the Siemens Excellence Award 2024 and the ZHAW Lab Sciences Award 2024.