“The AI Act was postponed” is only half true. The heavy part, the rules for high-risk systems, moved later. The rules that touch chatbots, voice assistants and AI-generated content stay in place, and those are the ones most small companies meet first.
What changed and when
Regulation (EU) 2026/1744 is dated 8 July 2026, was published in the Official Journal on 24 July 2026 and entered into force on 27 July. It is known as the Digital Omnibus on AI and amends the AI Act, Regulation (EU) 2024/1689 (Council corrigendum with the title and publication date, September 2026; Cuatrecasas, July 2026).
The new dates, as summarised by Cuatrecasas:
| What | New date |
|---|---|
| High-risk systems under Annex III (for example hiring and worker evaluation) | 2 December 2027 |
| High-risk systems that are safety components of regulated products (Annex I) | 2 August 2028 |
| New bans: non-consensual sexual deepfakes and child sexual abuse material | 2 December 2026 |
| Marking of AI-generated content for systems already on the market before 2 August 2026 | 2 December 2026 |
What is postponed, and whether it concerns you
High-risk is a narrow list. One entry is employment: AI for recruitment and CV screening, promotion and termination decisions, task allocation by behaviour, and monitoring workers (Cuatrecasas, 2026). If your company only uses AI to answer customers, book appointments or summarise e-mail, you are normally outside it. If you screen CVs with AI, you have until 2 December 2027, which is not long.
The Omnibus also adds relief for small companies and small mid-caps: specific definitions, simplified technical documentation and proportionate quality management (Cuatrecasas, 2026). That is for providers of high-risk systems, meaning the party that builds and places the system on the market, not for a company that merely uses a chatbot.
What already applies or starts this year
Telling people it is an AI. Article 50 requires systems that interact directly with people to be designed so that the person is informed they are dealing with an AI, unless that is obvious to a reasonably well-informed person. Article 50 applies from 2 August 2026 (AI Act Explorer, Article 50). A website chatbot and a voice assistant that answers calls should say so at the start.
The duty falls on the provider, the party that builds the assistant. If you use an off-the-shelf product, check that the supplier built the disclosure in, and do not remove it from your own script or widget.
Marking AI-generated content. Systems that generate text, images, audio or video must mark outputs in a machine-readable way (Article 50(2)). For systems already on the market before 2 August 2026 the deadline is 2 December 2026 (Cuatrecasas, 2026). This mostly concerns the providers of generative tools, not the companies using them.
Deepfakes and public-interest text. If you publish deepfakes, you must disclose that they are artificially generated. The same goes for AI-generated text published to inform the public on matters of public interest, unless a human with editorial responsibility has reviewed it (AI Act Explorer, Article 50). Ordinary marketing copy is not in that category.
AI literacy. After the amendment, Article 4 requires providers and deployers to take measures to promote AI literacy among staff who work with AI systems. It is a duty of means, not a guaranteed skill level, and the Commission will publish practical examples, especially for SMEs (Cuatrecasas, 2026).
A short checklist
- List where you use AI: website chatbot, phone assistant, e-mail summaries, content tools. Write down the supplier and your role for each.
- Check disclosure. The chatbot says at the start that it is an AI. The phone assistant says it in the first sentence. If you cannot verify it, ask the supplier.
- Keep a route to a person. Article 50 does not require it; this is our recommendation, because disclosure alone tells the customer about the problem without solving it. See our article on AI chatbots in business.
- Review what data the assistant receives. That is GDPR, not the AI Act: where data is processed, who the processor is, whether there is a contract. In our own AI work we prefer keeping data in the EU.
- Write a one-page AI use rule: who may use which tools and what must not be pasted in. It is also the simplest way to show you act on the literacy duty.
- If you screen CVs or evaluate staff with AI, plan a review before autumn 2027.
How to decide what to do now
If your AI talks to customers, fix disclosure and the hand-off to a person now; it is already in force. If AI works only internally, an inventory and a short rule are enough. If AI influences decisions about people, start with a legal assessment: the deadline is closer than it looks.
We build AI automation and assistants and the phone assistant DeskVoice, where disclosure and hand-off to a person are designed in from the start. If you want to check where AI makes sense in your processes and where it does not, discuss your project. On rolling out a phone assistant, see AI phone assistant instead of a call centre.
This article is not legal advice. For your own situation, ask a lawyer.
