EU AI ACT · TRANSPARENCY · ARTICLE 50
While everyone watched the EU delay its big AI rules, one part quietly stayed live - and it already applies to Swiss companies with EU customers.
Since 2 August 2026, the EU AI Act's transparency rules (Article 50) are in force - and they were not delayed by the October 2026 AI Omnibus. If your AI system interacts with people in the EU, or its output reaches the EU, three duties apply: disclose AI interaction, label AI-generated content, and disclose emotion-recognition or biometric categorisation.
The three duties in practice
Tell people they are talking to an AI. Chatbots and voice systems must disclose that the user is interacting with a machine, unless it is obvious. For a Swiss company with EU customers, that means the website chatbot says what it is.
Mark AI-generated content. Synthetic text, images, audio and video that resemble real people, places or events must be labelled as AI-generated. Machine-readable marking is required for synthetic content; a transitional deadline for systems already on the market before 2 August 2026 runs to 2 December 2026.
Disclose emotion recognition and biometric categorisation. People subjected to these systems must be informed.
Who this hits in Switzerland
Any Swiss company whose chatbot, voice assistant or AI content reaches EU users - no EU office needed; the Act follows the market, not the postcode.
Marketing and communications teams publishing AI-assisted imagery or video at scale.
HR and service teams using AI that reads emotion or categorises people (rare in SMEs, common in security and research contexts).
What still has time
The heavy machinery - conformity assessments and documentation for high-risk systems - was postponed by the AI Omnibus: 2 December 2027 for stand-alone high-risk systems, 2 August 2028 for AI embedded in regulated products. Transparency is the part that is already live.
The takeaway: the heavy rules were delayed - transparency was not. If your AI talks to EU users, it already has to say so.
The honest cost of ignoring it
The penalty regime has applied since August 2025: up to the higher of EUR 35 million or 7% of worldwide annual turnover for the gravest breaches, with lower ceilings for lesser ones - enforced by EU member-state authorities.
Sources
Council of the EU, "AI Act timeline"
EU AI Act Service Desk, "Timeline for the Implementation of the EU AI Act"
European Parliament Legislative Train, "Digital Omnibus on AI"
BSI, "EU AI Act: AI Omnibus Adopted" (13 October 2026)
Last updated: 20 October 2026.
TL;DR: The EU AI Act's transparency duties (Article 50) have applied since 2 August 2026 and were not delayed by the AI Omnibus. Swiss companies whose AI reaches EU users must disclose AI interactions, label AI-generated content, and disclose emotion recognition. The high-risk machinery moves to 2027-2028; transparency is already live, and fines apply.
Reading time: ~3 min
Not sure whether the transparency duties reach your product? We map that for Swiss companies in one working session. Discuss a project: eli5a.com/contact
