In short
- The Digital Omnibus postpones the rules for high-risk AI to December 2027 and August 2028. The transparency rules in Article 50 were not postponed and have applied since 2 August 2026.
- For marketing and content teams, that means: make chatbots recognisable as AI, label deepfakes, and make sure AI text gets human editorial review.
- AI literacy remains an obligation, but a lighter one: organisations must support it rather than guarantee a level.
What does the Digital Omnibus change in the EU AI Act? In short: a delay for high-risk AI, no delay for transparency. The Omnibus entered into force on 27 July 2026 (European Commission, 2026). The rules for systems that, for example, select job applicants or assess creditworthiness move to December 2027. The rules in Article 50, which make sure people know when they are dealing with AI, have simply applied since 2 August 2026.
For most marketing and content teams, Article 50 is therefore the part of the AI Act that matters now. Below you can read what was postponed, what Article 50 asks of you, when you need to label AI content and what to arrange this quarter.
This article is not legal advice. It is our translation of the rules into the day-to-day work of marketing and content teams. If you are unsure about your own situation, check with a lawyer.
What the Digital Omnibus postpones
The AI Act is the European law on artificial intelligence. The Omnibus is a package of simplifications the European Commission proposed in November 2025. The delay mainly concerns high-risk AI:
- High risk in applications (Annex III): from 2 August 2026 to 2 December 2027. Think of AI that selects job applicants, assesses creditworthiness or decides on access to education.
- High risk in products (Annex I): to 2 August 2028. This covers AI in products already subject to European safety rules, such as machinery and toys.
The Omnibus also extends a number of simplifications for SMEs to small mid-cap companies, and introduces a ban on AI systems that generate non-consensual sexual imagery or child sexual abuse material (European Commission, 2026).
Do you work in recruitment, finance or healthcare and use AI to assess people? Then you now have more time, but not less work. Law firm Gibson Dunn sums it up as a deferral, not a dismantling: the structure and core obligations of the law remain in place (Gibson Dunn, 2026).
What was not postponed: Article 50
Article 50 is about transparency. It contains three obligations that matter for marketing (AI Act, Article 50):
1. A chatbot identifies itself. When a visitor talks to an AI assistant, it must be clear that it is AI, unless that is already obvious to an attentive visitor. Formally, this obligation lies with the provider of the system. If you had the chatbot built and deploy it under your own name, that may be your organisation.
2. AI content gets a machine-readable marking. This is an obligation for the makers of AI tools, such as OpenAI, Google and Anthropic, not for you as a user. They must add a watermark or metadata to image, audio, video and text. Systems already on the market before 2 August 2026 have until 2 December 2026. That is the date you see in the news.
3. Deepfakes get a label. This one is an obligation for you as a user. The law defines a deepfake as image, audio or video generated or manipulated with AI that resembles existing people, objects, places or events and would falsely appear to be authentic. Think of a real person saying something they never said. If you create something like that, you disclose that it was made with AI. For evidently creative, satirical or fictional work, you can do so in a way that does not spoil the work.
Do you need to label AI content?
The short answer: not all AI content, but deepfakes and AI text on matters of public interest without human editorial review, yes.
- A video with an AI voice or AI avatar of a real person: label it.
- A product photo you made with AI that looks like a real photograph: this is a grey area, because the definition also covers objects. Label it if the image comes across as a real photo; that is the safe choice.
- A clearly drawn or stylised illustration: it does not falsely appear real, so a label is not required.
- A blog or newsletter you wrote with AI: a label is only required if you are informing the public on matters of public interest and there is no human editorial review with someone holding editorial responsibility. Marketing copy that a colleague has read and approved generally falls outside this.
- A chatbot on your website or in customer service: make it visible that it is AI.
The European Commission encourages codes of practice for marking and labelling AI content, so the details may still be filled in further. Keep an eye on the guidance if you work with AI images or video a lot.
Is AI literacy still mandatory?
Yes, but lighter. Since 2 February 2025, organisations have had to ensure that staff working with AI understand it sufficiently (Article 4). The Omnibus has softened this: organisations must now support the development of AI literacy rather than guarantee a particular level. The European Commission and member states take on a larger role in promoting it.
The reason to train your team has not changed. An employee who does not know what a model can and cannot do pastes customer data into a public tool or publishes copy with a made-up figure. That risk does not depend on a legal article.
What to arrange this quarter
- Check your chatbots. Does it say that it is AI? If not, add it.
- Set a house rule for AI images and video. When do you label, and how? One sentence in your brand guidelines is enough to start.
- Decide who reviews AI copy. Human editorial review is not only good for quality. It is exactly the exception the AI Act names.
- Know which AI tools your team uses. A simple list: tool, what for, and with which data. You need it for every next step.
- Do you use AI in recruitment, credit or healthcare? Then start preparing for December 2027 now, not in November.
How we help
In our workshops your team learns to work with AI, including safe use, privacy and the limits of a model. You receive an attendance list that shows who has been trained. In an AI workplace we capture the agreements: which tools, which data, who reviews. Then they do not just sit in a policy document; they are built into the work itself.
Want to know what the AI Act means for your organisation in practice? Book a free conversation.
Also read Using AI agents safely: five questions before you start and ChatGPT or Claude? With GPT-6 and Opus 5.5, that is the wrong question.
Sources
- European Commission (27 July 2026). AI Omnibus enters into force.
- Regulation (EU) 2024/1689 (AI Act), Article 50: transparency obligations.
- Gibson Dunn (27 May 2026). EU AI Act Omnibus Agreement: Postponed High-Risk Deadlines and Other Key Changes.
