Insights · AI Governance

EU AI Act: the delay is not a pause

The EU has postponed the obligations for high-risk systems to the end of 2027. At the same time, transparency obligations that affect almost every organisation have applied since August 2026. And Switzerland is preparing its own framework.

In June 2026, the Council of the EU gave final approval to the so-called Digital Omnibus on AI. The most important change: the obligations for stand-alone high-risk systems under Annex III no longer apply from August 2026, but from 2 December 2027. For AI in products that are already regulated under Annex I, such as medical devices or machinery, the date is 2 August 2028.

Since then, I have heard the same sentence in many conversations: the AI Act has been postponed. That is a dangerous oversimplification.

What has moved, and what has not

What has moved is the timing of the high-risk obligations, not their content. Conformity assessment, quality management, technical documentation and registration remain the same. Anyone operating a system today that falls into the high-risk category, such as candidate screening, performance assessment or credit scoring, has more time, but no lighter requirements.

Everything else remains in force unchanged. The prohibitions have applied since February 2025, as has the obligation to ensure sufficient AI literacy among staff. And since August 2026, the transparency obligations under Article 50 apply: people must be told when they are interacting with an AI system such as a chatbot, and deepfakes and AI-generated text on matters of public interest must be labelled.

The deadline for the heavy obligations has moved. The obligations that affect almost every organisation already apply.

What this means in practice for SMEs

For most mid-sized organisations, the immediate issue is not the high-risk category but transparency. The chatbot on the website, the AI assistant in customer service, automatically generated text in newsletters or on social media: all of this falls under Article 50. Implementation is usually simple, a clear notice or a label. But someone has to own it and check it.

And Switzerland?

Switzerland has no AI law of its own and does not plan to copy the AI Act. By signing the Council of Europe's Framework Convention on AI, however, it has committed to adapting its own law. According to the federal SME portal, a consultation draft is due in spring 2027, covering transparency, data protection, non-discrimination and oversight in particular.

Until then, existing law applies, above all the revised Federal Act on Data Protection. And Swiss companies whose AI systems or their outputs have an effect in the EU are directly subject to the AI Act, including the transparency obligations.

How to use the time you have gained

  • Review the transparency obligations now. Where do customers or applicants interact with AI, and is that clear to them?
  • Keep the register of your AI systems up to date. It is the basis for any classification, in the EU and in Switzerland alike.
  • Use the time until December 2027 for high-risk systems to build proper documentation and human oversight, rather than pausing the work.
  • Document your employees' AI literacy. This obligation has applied since February 2025 and is the first thing any audit will ask about.

Those who treat the delay as a pause will find themselves under time pressure at the end of 2027, at the same moment as everyone else, with limited capacity at assessment bodies and advisers. Those who use it as room to plan will get the work done calmly.

Sources

Invitation

Let us discuss this in your context

Articles stay general. Your situation is not. In sixty minutes we translate the thinking to your starting point.

Book a sparring session

No preparation, no obligation.