In early 2026, the Digital Omnibus softened part of the EU AI Act. One message dominated the business press: the obligations for high-risk AI are moving to the end of 2027. In many boardrooms, that landed as relief. August 2, 2026 puts a question mark behind it. On that day, the obligations that affect companies in Germany across the board take effect. That leaves less than two weeks.
What Actually Takes Effect on August 2, 2026
Two things start on this date, and both go beyond the makers of high-risk systems.
The transparency obligation under Article 50. Anyone using AI must, from August 2, 2026, disclose where people are interacting with an AI.1 Chatbots have to identify themselves as AI. AI-generated images, video, or audio must be labeled as such and marked in a machine-readable format. Deepfakes and AI-generated text published to inform the public must be disclosed. For generative systems already on the market before August 2, 2026, a transition period until December 2, 2026 applies for the machine-readable marking.1
Enforcement against providers of general-purpose AI models (GPAI). From August 2, 2026, the European Commission can actively enforce the GPAI obligations through the AI Office: request documentation, technically evaluate models, demand remediation, and impose fines.2 The ceiling for GPAI infringements is up to 3 percent of worldwide annual turnover or 15 million euros, whichever is higher.2 The higher figures of up to 35 million euros or 7 percent of turnover apply to prohibited practices, not to the ordinary case in a company.
What Was Postponed, and Whom It Affects
The Digital Omnibus moved the obligations for standalone high-risk AI systems under Annex III to December 2, 2027.3 AI components in regulated products such as medical devices get even more time. The reason is practical: conformity assessment for high-risk systems is demanding, and the technical standards for it are not yet finished.
This postponement affects few companies directly. Anyone who does not build AI for hiring decisions, credit scoring, or comparable fundamental-rights-relevant decisions rarely falls under Annex III. The transparency obligation, by contrast, hits almost everyone who uses AI in customer contact or content production. That is exactly the obligation that stays on August 2, 2026.4 For a full overview of all obligations and deadlines, see our guide to the EU AI Act for businesses.
Why Companies in Germany Should Act Now
Adoption is outrunning control. 40 percent of the German economy now uses AI, up 118 percent since 2024.5 The expected return is rising with it: a SAP/Oxford study puts the AI ROI of German companies for 2026 at 24 percent. The same study notes that governance is lagging behind.6 A readiness survey earlier this year found 78 percent of companies unprepared for the EU AI Act.7
The fine risk is real, but manageable. It comes down with structure, not with a scramble in the final weeks before the deadline. The decisive question is the same for almost every company: which AI systems are running here, which risk class do they fall into, and who is responsible? This groundwork depends on no postponed deadline. It is the precondition for meeting the August obligations without panic.
Four Steps Before August 2
- Build an inventory. Get an overview of which AI systems are in use, including the features quietly built into existing software. Without an inventory, no classification is possible.
- Assign risk classes. Place each system in a category: prohibited, high-risk, subject to transparency, or minimally regulated. Most applications in German companies fall into the lower categories. Prove it, rather than assume it.
- Prepare transparency and labeling. Clarify where users need to recognize that they are talking to an AI, and where AI-generated content is produced that has to be marked. At the same time, check your contracts with providers to see whether they meet their labeling obligations.
- Assign responsibility. Name a role for AI governance. Without clear ownership, every measure stays a one-off project instead of an ongoing process.
These four steps are not an end in themselves. They are the basis for scaling AI in a company with legal certainty. Whoever knows which systems they use, and for what, also makes better decisions about where AI pays off.
Conclusion
The Digital Omnibus moves the complex high-risk obligations to 2027. That relieves pressure. The obligations that affect companies in Germany across the board take effect on August 2, 2026. The right response to the headline "deadlines postponed" is to start a structured inventory. Legal certainty comes from a framework set up in good time, not in the last weeks before a deadline.
Sources & References
- European Commission / AI Act: Article 50 – transparency obligations, applicable from August 2, 2026 (machine-readable marking: transition until December 2, 2026): artificialintelligenceact.eu
- European Commission, Guidelines for providers of general-purpose AI models – enforcement and fines from August 2, 2026 (up to 3% of turnover / €15 million): digital-strategy.ec.europa.eu
- Börse Express / Council of the EU, June 30, 2026: EU postpones high-risk rules until December 2027 (Digital Omnibus): boerse-express.com
- heise online, June 11, 2026: EU AI Act – what needs to be done in companies by August 2026: heise.de
- Handelsblatt (IW Consult / eco), July 8, 2026: 40% of the German economy uses AI, +118% since 2024: handelsblatt.com
- Computerwoche, July 15, 2026: SAP/Oxford study – AI ROI of German companies 24%, governance lagging behind: computerwoche.de
- Vision Compliance / National Law Review, April 2, 2026: 2026 EU AI Act Readiness Report – 78% of companies unprepared: natlawreview.com