Deferred is not suspended

Today is 2 August 2026, the general date of application for the EU AI Act. Over the coming days many posts will state that the AI Act applies from today. That is true but misleading, because the content of this date changed five days ago.

What happened at the end of July

On 24 July, Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published in the Official Journal. It entered into force on 27 July. This is not an announcement or a draft but an amending regulation that has been in force for a week. It defers the obligations for high-risk systems under Annex III from today to 2 December 2027. For high-risk AI in regulated products under Annex I, the date moves to 2 August 2028.

The reason is unspectacular and telling at once: the harmonised standards and the national supervisory authorities are not ready. It is hard to demand conformity against standards that do not yet exist.

Who is relieved by the deferral, and who is not

This is where it gets interesting for most organisations. The high-risk classification under Annex III affects comparatively few, covering eight clearly delimited areas such as biometrics, critical infrastructure, employment or access to essential services. Anyone outside those was never affected by today's date to the degree the excitement suggests.

The transparency duties under Article 50, by contrast, were not deferred, and they affect almost anyone using AI in external contact. Whoever runs a chatbot must disclose that the other party is talking to an AI system, unless that is obvious anyway. Whoever generates or alters images, audio or video with AI must mark it in machine-readable form. The AI literacy duty under Article 4, in force since February 2025, likewise remains unchanged.

One qualification belongs with this, and it comes from the same amending regulation. It adds a new Article 111(4) to the AI Act: providers of AI systems that generate synthetic audio, image, video or text content, and that placed those systems on the market before today, must comply with the machine-readable marking under Article 50(2) by 2 December 2026. This is to be read narrowly. The deadline applies to providers and their procedures, not to retroactively labelling content published at some earlier point. Anyone first offering such a system from today does not have it at all. And the chatbot disclosure duty is unaffected by it.

That inverts the situation: those who breathed a sigh of relief at the deferral are often the ones who have to act from today. And those who genuinely operate high-risk systems did not get a break but an extension of the preparation window. Reading it as a reprieve loses the time twice over.

The uncomfortable part

The first question is not which duty applies from when. It is: where in the organisation is AI actually being used? In most organisations nobody knows completely, because AI does not arrive as a project but as a feature in tools that are already there. Without that list every classification is guesswork, and then a sixteen-month deferral does not help either.

Where isidaten comes in

At exactly that list. AI management maintains the provider integrations, the AI users and the mapping of which function may use which access, including activity logs. The legal register maintains obligations at clause level and includes an editorial catalog for the AI Act, so a deferred deadline does not vanish as a footnote. And the ISO 42001 solution page describes the management system for running this over time.

Product-related context, not legal advice. The text of the regulation governs: the amending regulation is published as Regulation (EU) 2026/1744 in the Official Journal.