Summary
- The Digital Omnibus on AI, adopted by the European Parliament on June 16, 2026, and subsequently by the Council of the EU on June 29, 2026, postpones the obligations related to “high-risk” AI systems to December 2027 (Annex III) and August 2028 (Annex I).
- This postponement does not apply to the transparency requirements of Article 50, which do take effect on August 2, 2026 for any chatbot, callbot, or voicebot that interacts with your customers.
- Virtually all B2B conversational AI systems (customer support, appointment scheduling, FAQs) fall under the “limited risk” category: they were never affected by the “high-risk” deferral and remain fully subject to Article 50.
- Penalties for failure to comply with transparency requirements can reach 15 million euros or 3% of global revenue.
- Three specific tasks remain to be completed before the deadline: clear user information, system documentation, and a mechanism for transferring the conversation to a human.
For the past few weeks, a notion has been circulating in marketing and legal departments: “The AI Act has been postponed; we have time.” This is both true and false. On June 29, 2026, the Council of the EU gave its final green light to the Digital Omnibus on AI, which does indeed postpone part of the regulation. But not the part that directly concerns your chatbot, your callbot, or your voicebot. Here’s what has actually changed, and what still needs to be done in a few days.
The new AI Act timeline, in a minute
The postponement voted on by the European co-legislators applies to “high-risk” AI systems as defined by the AI Act (automated recruitment, credit scoring, medical devices, critical infrastructure, etc.). These requirements, which were originally set to take effect on August 2, 2026, are now set for December 2, 2027 for autonomous systems (Annex III) and August 2, 2028 for AI integrated into regulated products (Annex I).
This is a real breath of fresh air for the affected sectors. However, conversational AI used in customer interactions—such as chatbots, callbots, and voicebots—has, in the vast majority of cases, never fallen into this category. The postponement therefore does not apply to it.
What Remains Unchanged as of August 2, 2026: Transparency of Your Conversational AI
Article 50 of the AI Act, which governs the transparency of conversational systems, remains in effect as of August 2, 2026, with no changes. Specifically, two requirements form the basis of compliance for any conversational system:
- Inform the user that they are interacting with an AI and not a human, from the very start of the interaction. This rule applies to chatbots, callbots, and voicebots alike: an introductory voice message played before a call has the same effect as a welcome message on a chat interface.
- Label content generated by AI (text, audio) in an identifiable format when the system relies on generative AI to produce its responses.
Something to keep in mind: The European Commission has provided a grace period, but only for labeling content from systems already on the market before August 2, 2026—which has been extended to December 2, 2026. The obligation to inform the user, however, has no grace period.
The pitfall to avoid: believing that “extended” means “no need to worry”
This is the most costly mistake to anticipate. A customer support chatbot, an appointment-scheduling callbot, or a call-qualification voicebot fall, in their standard use, under the “limited risk” category of the AI Act. They have therefore never been subject to “high-risk” obligations, and the postponement voted on in late June does not change their status.
Another point to watch out for: as your chatbots and callbots evolve into AI agents capable of acting autonomously (initiating a refund, modifying a reservation without human validation), their classification could shift to a higher risk level. The postponement of the “high-risk” classification to 2027/2028 is therefore not a long-term guarantee for the most autonomous use cases.
What the August 2 deadline does change, however, is the full activation of the penalty regime. A breach of the transparency requirement can now cost up to 15 million euros or 3% of the company’s global annual revenue, whichever is higher.
Quick checklist: Is your conversational AI ready?
- Information visible from the very first interaction, across all your channels: website, app, social media, and for a callbot or voicebot, as soon as the call begins.
- Labeling of content generated by your generative AI system, in a readable and detectable format.
- Documentation and traceability of the system, with a clear classification of risk levels based on use cases.
- Mechanism for switching to a human agent, accessible and seamless, in addition to meeting legal requirements.
How Dydu Ensures Your Compliance Across All Channels
At Dydu, compliance isn’t a separate project: it builds on existing product choices across the entire conversational channel—chat, call, and voice.
Our solutions are hosted in France, which simplifies GDPR compliance—an essential prerequisite for any AI Act compliance effort. This hosting in France is also part of a digital sovereignty strategy: your data and that of your customers remain beyond the reach of non-European extraterritorial regulations, an issue increasingly scrutinized by legal and procurement departments. The platform allows you to configure the transparency message as soon as a conversation begins—whether written or voice-based—without requiring any specific development. And the seamless handoff to a human agent is a core principle of our approach: the bot assesses the request and then transfers it according to configurable rules, without disrupting the user experience.
Share this article!
