The EU AI Act Hasn't Been Delayed: What Actually Changes on August 2, 2026
The EU AI Act has not been delayed. One specific part of it has: the high-risk obligations under Annex III (hiring, credit scoring, educational assessment, and similar use cases), which move from August 2026 to December 2027. Article 50 — the requirement that every chatbot or voice agent disclose that it is AI — has not moved. It still takes effect on August 2, 2026, exactly as originally scheduled.
The confusion has a clear source. The Digital Omnibus, formally adopted by the European Parliament on June 16, 2026 and by the Council of the EU on June 29, has been read by much of the market as a blanket postponement of the regulation. It isn't. It's a surgical adjustment to a single risk category.
What the Digital Omnibus actually changed
| Date | Milestone | What it affects |
|---|---|---|
| June 16, 2026 | Formal adoption by the European Parliament | Digital Omnibus approval |
| June 29, 2026 | Formal adoption by the Council of the EU | Entry into force after publication in the Official Journal |
| August 2, 2026 | Article 50 takes effect | AI disclosure for chatbots and voice agents, labeling of deepfakes and synthetic content — unchanged |
| December 2, 2027 | New deadline for Annex III | High-risk system obligations (hiring, credit, education) — delayed 16 months |
A customer-facing conversational agent is, outside a few narrow exceptions, not an Annex III high-risk system. It's a system subject to transparency obligations under Article 50. Its deadline is still August 2, 2026.
Why this matters to a CX leader or COO right now
If your company operates a chatbot, voice agent, or any conversational system with customers in the EU, two concrete obligations land on August 2, 2026: the agent must disclose it is AI during the conversation itself — not buried in a footer or terms page — and any synthetic or deepfake content it generates must be labeled. Neither obligation moved. Assuming "everything got delayed" and deprioritizing the review of your agent is the mistake that will cost some companies in September.
Frequently asked questions
Has the EU AI Act been delayed?
Not in its entirety. Only the high-risk obligations under Annex III were delayed, moving from August 2026 to December 2027. The rest of the regulation, including Article 50, keeps its original timeline.
What is the Digital Omnibus?
It's the legislative package the European Commission introduced to simplify and adjust deadlines across several digital regulations, including the EU AI Act. It was formally adopted by the European Parliament on June 16, 2026 and by the Council on June 29, 2026.
Does the delay apply to chatbots and voice agents?
No. Chatbots and voice agents fall under Article 50 (transparency), not Annex III (high-risk). Their obligation to disclose AI status still takes effect on August 2, 2026, unchanged.
What obligations take effect on August 2, 2026?
Explicit disclosure that a conversational system is AI from the first exchange, and labeling of synthetically generated or manipulated content — deepfakes, synthetic voices, AI-generated images.
What happens if my company doesn't comply with Article 50?
EU AI Act penalties can reach €15 million or 3% of global annual turnover. But regulatory exposure isn't the only risk, or even the most likely one: an agent that doesn't disclose it's AI and then fails mid-conversation creates a trust and reputational problem with the customer well before any regulator gets involved.
The calendar mix-up that gets expensive
Key insight from LEXIC.AI: When a company confuses "part of the regulation was delayed" with "the regulation was delayed," it spends its remaining runway on something that doesn't apply to it, and arrives at August 2 without having reviewed what actually did. The date didn't change. The perception of the date did.
"The Digital Omnibus was designed to give high-risk systems breathing room. It never touched conversational transparency. Confusing the two is exactly the kind of regulatory misread that good AI governance should have caught weeks ago — not the week before the deadline," says Sergio Llorens, CEO of LEXIC.AI.
How to know if your agent is ready for August 2
Governing a conversational AI agent well isn't a last-minute patch. It's having independently verified evidence, drawn from real production conversations, that the agent discloses correctly, escalates when it should, and never generates unlabeled synthetic content. An independent Lexic Compass report delivers that verdict across 100% of production conversations — not a sample, and not a test environment.
If you want to know exactly where your agent stands before August 2, book a demo with your own data.