Regulation laying down harmonised rules on artificial intelligence (EU AI Act).

When does EU AI Act apply to you?

EU AI Act may apply to you if one of these triggers applies. The Regulatory Check tests them against your details.

• Activity: AI provider (own AI products) (Likely applies): As the provider of an AI system placed on the market in the EU, you bear the most far-reaching obligations under the EU AI Act - from bans on certain practices to conformity assessment and transparency obligations, depending on your system's risk class.
• Activity: Using AI (Recommend individual review): Even as a deployer - not only as a provider - of an AI system, you have obligations under the EU AI Act, in particular on transparency towards users and, for high-risk systems, on oversight and documentation.
• Markets: EU (applies if additionally: Activity: Using AI, AI provider (own AI products)) (Recommend individual review): The EU AI Act is triggered by placing on the market or use in the EU, not by the registered seat - if you use AI systems whose output is used in the EU, the regulation can apply regardless of your seat in Switzerland. Without any use or provision of AI, the EU market alone does not trigger the regulation.

Exceptions

• AI systems developed or used exclusively for military, defence or national security purposes.
• Purely private, non-professional use of AI systems by natural persons.
• Scientific research and development prior to market introduction.
• Systems that, per the Digital Omnibus amendment (Regulation (EU) 2026/1744), serve exclusively user support, performance optimisation, automation or non-safety-relevant quality control no longer automatically fall under the high-risk classification.

What does EU AI Act require?

• Refrain from prohibited AI practices (including manipulative systems, social scoring, certain real-time remote biometric identification) since 2 February 2025; two further bans (including on non-consensual intimate content and abuse material) have applied since 2 December 2026.
• For providers of general-purpose AI (GPAI) models: transparency and copyright obligations, plus additional risk assessments where there is systemic risk, since 2 August 2025.
• Transparency obligations under Art. 50: labelling of AI-generated content, disclosure of interaction with an AI system, labelling of deepfakes and AI-generated text on matters of public interest - since 2 August 2026, with a four-month transition period for systems already on the market until 2 December 2026.
• For high-risk AI systems under Annex III (including biometrics, critical infrastructure, education, employment, migration/asylum/border control): conformity assessment, risk management system, documentation, human oversight - applicable from 2 December 2027.
• For high-risk AI as a safety component in regulated products under Annex I (e.g. lifts, toys): corresponding obligations from 2 August 2028.

What evidence is needed?

• Risk classification of the AI system (prohibited/high/limited/minimal) with justification.
• Technical documentation and declaration of conformity for high-risk systems.
• Labelling or disclosure under Art. 50 where transparency obligations apply.
• Documentation of human oversight for high-risk use.

Which deadlines apply?

1. 2 February 2025 Bans on certain AI practices and the obligation on AI literacy become applicable.
2. 2 August 2025 Rules for general-purpose AI (GPAI) models become applicable.
3. 24 July 2026 The amending regulation (EU) 2026/1744 ('Digital Omnibus') is published in the Official Journal.
4. 2 August 2026 Transparency obligations under Art. 50 become applicable.
5. 2 December 2026 The transition period for transparency obligations for systems already on the market ends; two further bans under Art. 5 enter into force.
6. 2 December 2027 Obligations for high-risk AI systems under Annex III become applicable (postponed from originally August 2026).
7. 2 August 2028 Obligations for high-risk AI as a safety component in regulated products under Annex I become applicable (postponed from originally August 2027).

Information as of: 24 September 2026. Past dates are grey, upcoming ones highlighted (as of when the page was built).

What are the consequences of violations?

Breaches of prohibited AI practices (Art. 5) can be penalised with fines of up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher. Breaches of transparency obligations under Art. 50 can be penalised with up to EUR 15 million or 3% of worldwide annual turnover.

What is connected with it?

• Swiss AI regulation (planned implementation of the Council of Europe Framework Convention on Artificial Intelligence): Switzerland does not yet have an AI-specific act.
• Regulation (EU) 2017/745 on medical devices: The EU Medical Device Regulation (MDR, Regulation (EU) 2017/745) governs conformity assessment, CE marking, technical documentation and market surveillance for medical devices in the EU; it has applied since 26 May 2021.

What is still uncertain?

• The EUR-Lex primary text of Regulation (EU) 2024/1689 and the amending regulation (EU) 2026/1744 could not be retrieved in this session (empty response, per prior findings HTTP 202/bot block); deadlines evidenced via the EU Commission page and a law-firm source, not via the regulation's text itself.
• The exact entry-into-force date of Regulation (EU) 2026/1744 is given in a secondary source as '27 July 2026' (publication per prior findings on 24.7.2026) - not checked against the Official Journal itself.
• The exact new wording of 'safety-relevant components' after the Omnibus amendment was not checked in full text.

Sources

Information as of: 24 September 2026.

• Regulatory framework proposal on artificial intelligence, Europäische Kommission (Generaldirektion CNECT), retrieved on 24 September 2026
• EU AI Act: Was ab dem 2. August 2026 gilt – und was verschoben wurde, Flick Gocke Schaumburg (Kanzlei), retrieved on 24 September 2026

A professional assessment based on publicly available sources, not legal advice. Whether a given rule applies in your specific case depends on circumstances that are not fully captured here.