August 2 Has Arrived – and a Misunderstanding About the 'Digital Omnibus' Came With It

Yesterday, August 2, the EU AI Act became, in principle, fully applicable under Article 113. At the same time, a shortened version of that news has been circulating for weeks: that the EU's so-called Digital Omnibus already pushed back the central obligations of the AI Act anyway, effectively defusing the deadline. That reading is widespread – and wrong, at exactly the point that matters most to most companies.

What the Digital Omnibus actually postponed

It's true that the EU reordered the AI Act's deadline structure with the Digital Omnibus and pushed the heavily debated obligations for high-risk AI systems – in areas like hiring, credit decisions, or critical infrastructure – back by 16 months. For companies using such high-risk systems, that's a genuine breathing room, with noticeable financial consequences: compliance costs for a single high-risk system alone are estimated at 50,000 to 500,000 euros for mid-sized companies, and 8 to 15 million euros in the first year for large enterprises.

What still applies – and why that gets missed

Explicitly not covered by that postponement is Article 50 – the labeling requirement for AI-generated content, which I already wrote about back in July. That obligation took effect yesterday, unchanged by the Digital Omnibus, and affects not just tech corporations but every company with customer-facing AI – which in practice means nearly every company publishing AI-generated images in marketing, running chatbots in customer service, or publishing AI-generated text.

The misunderstanding arises because both pieces of news – "16 months of delay" and "deadline August 2" – circulate at the same time and rarely get cleanly separated. Whoever only reads the headline about the Digital Omnibus easily assumes the entire AI Act has been defused. That exact confusion is currently the bigger risk than the regulation itself.

Why this matches exactly what I see in consulting

This blurriness confirms a pattern that runs through practically every current study on AI compliance: most companies have no formal inventory of their AI systems, most have nobody internally responsible for compliance, and most have no process to even produce the required technical documentation. That matches exactly what I wrote about the study on failed AI projects: the technology isn't the problem – the missing stocktaking is. Whoever doesn't even know which AI systems are running inside their own company can neither assess whether they fall under the postponed high-risk obligations, nor whether Article 50 already applies to them today.

What to do now

The pragmatic first step stays the same one I teach in our AI seminars for companies, just with an updated order: first check whether your own AI applications fall under Article 50 – that already applies now to most companies with marketing or customer-facing AI, regardless of the Digital Omnibus. Only after that should you settle the question of whether you're even running high-risk systems in the sense of the regulation, for which there's now more time. Whoever reverses that order wastes the extra time on the high-risk question while overlooking the obligation that already applies.

Conclusion

The Digital Omnibus is genuine relief – but only for part of the regulation. Whoever turns that into a general all-clear is confusing a postponed deadline with a canceled one. Article 50 has applied since yesterday, regardless of how much extra time was granted elsewhere – and making that distinction cleanly is exactly the real compliance task right now.