One day, two pieces of news

Some coincidences work as an image. On 24 July 2026, two things happened that have nothing to do with each other and yet say everything about this year.

• Anthropic released Claude Opus 5 — the next top model, eight months after Opus 4.5.
• The EU's Official Journal carried Regulation (EU) 2026/1744, which postpones the high-risk obligations of the AI Act to December 2027.

Both are documented and can be read up on — here is the Official Journal text: Regulation (EU) 2026/1744 on EUR-Lex.

Two clocks running at different speeds

Between the Commission's proposal on 19 November 2025 and its entry into force on 27 July 2026 lay a good eight months. In the same period — by our verified count — at least fifteen new models appeared from Anthropic, OpenAI and Google.

A legislative procedure takes longer than four model generations.

That is not a criticism of the EU. A law has to pass through two chambers, through translation into 24 languages, through legal scrutiny. That this is slower than a product release goes without saying — it only stands out when you place the two side by side.

What a business makes of it

The obvious conclusion would be: do not wait for the rules, they arrive too late. That one is wrong. The prohibitions in Article 5 have applied since February 2025, the transparency obligation since 2 August 2026, and the Swiss Data Protection Act has applied to AI for a long time anyway.

The useful conclusion is a different one: build in such a way that neither clock catches you out. Concretely that means — keep the model interchangeable, so that a switch costs hours rather than weeks. And build traceability in from the start instead of waiting for the date on which it is demanded.

Anyone who logs who saw and changed what and when is not thereby meeting some future rule. They know what has happened in their business. That was a good reason before the AI Act and remains one, however often the deadline is postponed.