AI Act from 2 August 2026: what your chatbot must tell customers
From 2 August 2026 a chatbot in the EU must say it is an AI. What the digital omnibus moved, what it left alone, and 6 steps for a small business.
Nikolay Tolmachev, Founder, Tolman Studios
6 min readNews
In short
2 Aug 2026the date chatbots in the EU must say they are AI. The digital omnibus did not move it
- From 2 August 2026 a chatbot must make clear that it is an AI, at the latest in its first message, unless that is obvious. Bots that are already running are included.
- The digital omnibus, Regulation (EU) 2026/1744 (in force 27 July), moved the high-risk rules to December 2027 and August 2028. It did not move the chatbot rule.
- Realistic AI photos, videos or voices of real people or places need an "AI-generated" label. A blog post that you checked and stand behind usually does not.
On this page
From Sunday, 2 August 2026, a chatbot in the EU must make clear to people that they are talking to an AI. This has to happen in the first message at the latest, unless it is already obvious. The new "digital omnibus" came into force on 27 July and did not move this date.
For most small businesses this means one extra line in the greeting. I'm not a lawyer. Below I explain what the law says, link the official texts and show what I do on my own site.
What applies from 2 August 2026?
Article 50 of the AI Act, Regulation (EU) 2024/1689, is about transparency. Three parts matter for a small business.
- Chatbots (Article 50(1)). An AI system that talks to people must be built so they know it is an AI. The exception is when that is obvious to a reasonably attentive person. A bot called "Jana" with a friendly photo is not obvious.
- Deepfakes (Article 50(4)). If you publish AI images, video or audio that look like real people, places or events and could pass as genuine, you must say they are AI-made. For clearly artistic or playful work a light mention is enough.
- AI text on public matters (Article 50(4)). AI text published to inform the public on matters of public interest needs a label. There is no label needed if a person reviewed or edited it and takes responsibility for it.
Article 50(5) says how to do it: clearly, at the first contact at the latest, and in an accessible way. The rule also covers bots that have been running for years.
In June 2026 the European Commission's AI Office published the final Code of Practice on marking and labelling AI-generated content. It is voluntary and written mainly for AI vendors. Its part for "deployers", the businesses that use AI, shows what good labels can look like.
What did the digital omnibus change, and what not?
The omnibus is a package of changes that simplifies the AI Act. The Council approved it on 29 June 2026. It was signed on 8 July, published on 24 July and came into force on 27 July. It moved the big deadlines for high-risk AI. The Article 50 dates stayed where they were.
| Rule | Before the omnibus | After Regulation (EU) 2026/1744 |
|---|---|---|
| Chatbot says it is AI (Art. 50(1)) | 2 Aug 2026 | 2 Aug 2026, unchanged |
| Labels for deepfakes and AI text on public matters (Art. 50(4)) | 2 Aug 2026 | 2 Aug 2026, unchanged |
| Machine-readable marking of AI output (Art. 50(2), the vendor's job) | 2 Aug 2026 | 2 Aug 2026 for new tools, 2 Dec 2026 for tools already on the market |
| Staff AI literacy (Art. 4) | Since 2 Feb 2025: "ensure a sufficient level" | Still applies: "take measures to support", no fixed level |
| High-risk AI in areas like hiring, education, credit (Annex III) | 2 Aug 2026 | 2 Dec 2027 |
| High-risk AI in regulated products (Annex I) | 2 Aug 2027 | 2 Aug 2028 |
Which dates matter?
1 Aug 2024
The AI Act comes into force.
2 Feb 2025
Bans on the worst AI uses apply. Article 4 on staff AI literacy applies.
2 Aug 2025
Rules for general-purpose AI models apply, the ones behind ChatGPT, Gemini or Claude. Countries must have their penalty rules ready.
27 Jul 2026
The digital omnibus, Regulation (EU) 2026/1744, comes into force.
2 Aug 2026
Article 50 applies: chatbots say they are AI, deepfakes get labels. National authorities can enforce it.
2 Dec 2026
Grace period ends for machine-readable marking by generative AI tools already on the market. New bans on nudify apps and AI child abuse material.
2 Dec 2027
High-risk rules for Annex III areas such as hiring, education and credit.
2 Aug 2028
High-risk rules for AI inside regulated products such as machines, toys and medical devices.
What should a business with a website chatbot do now?
- Say it in the first message. For example: "Hi, I'm the AI assistant of Salon Jana. I answer from our price list." Put "AI assistant" in the chat window title too.
- Show a way to reach a person. A button, or a line like "Write 'person' and Jana replies". The law doesn't ask for this directly. Customers expect it, and it saves you angry reviews.
- Check every channel. The same line belongs in your WhatsApp, Instagram or phone bot. A voice bot says it at the start of the call.
- Ask your vendor. If you use a ready-made chatbot tool, ask whether the AI notice is on by default and whether you can edit it.
- Brief your staff. One hour is enough. Cover what the bot answers, where it goes wrong and what never to paste into AI tools. Write down the date and who was there. This is your Article 4 homework, and it has been due since February 2025.
- Label realistic AI pictures. An AI "photo" of your shop or of a smiling customer in an ad gets a short "AI-generated" note.
On my own site the chat window says "AI assistant · answers from this website" under its title. The first message starts with "Hi! I'm the AI assistant here." Under the chat there are buttons to write to me directly on WhatsApp, Telegram or by email. A short line also says that AI can make mistakes. I set up bots for clients the same way.
How big are the fines?
- maximum fine for breaking Article 50, the higher of the two, for large companies
- €15M or 3%
- AI Act, Art. 99(4)
- for SMEs and start-ups the cap is whichever of the two is lower
- the lower one
- AI Act, Art. 99(6)
- the cap for a firm with €200,000 yearly turnover (3%)
- €6,000
- my calculation from Art. 99(6)
So for a small firm the ceiling is 3% of last year's worldwide turnover. The omnibus also added warnings and other measures that are not fines to Article 99. Authorities must consider whether a small firm can survive the fine. Each country sets the details. Nobody knows yet how often they will check small websites. One line in your greeting costs nothing.
What should you read next?
Still deciding whether a chatbot is worth it? Read where AI chatbots help and where they don't. If you use Meta's free agent on WhatsApp, my post on Meta Business Agent covers the greeting there. If you want me to check your bot's first message or set one up properly, see what I do in AI and automation.
Questions people ask
Did the digital omnibus postpone the chatbot rule?
No. Article 50 still applies from 2 August 2026. The only grace period is for machine-readable marking of AI output (Article 50(2)), and only for generative AI tools already on the market before 2 August. They have until 2 December 2026. That marking is the AI vendor's job.
Does this apply to a WhatsApp or Instagram bot too?
Yes. The rule covers any AI system that talks to people, wherever it runs. Start the conversation with one line that says it is an AI assistant.
Do I have to label blog posts I write with ChatGPT?
Usually not. The label is for AI text published to inform the public on matters of public interest, and even then not if a person reviewed or edited it and takes responsibility for it.
How big are the fines?
Up to €15 million or 3% of worldwide yearly turnover. For small and medium firms the lower of the two counts. For a firm with €200,000 turnover that caps the fine at €6,000.
Do my staff need an AI course?
Since 2 February 2025, Article 4 has asked firms that use AI to look after their staff's AI literacy. The omnibus made it softer. You take measures to support it, and nobody expects a fixed level. A short briefing that you record in writing is a good start.
Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 4, 50 and 99, EUR-Lex, checked 28 July 2026
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L of 24 July 2026, EUR-Lex, checked 28 July 2026
- AI Omnibus enters into force, European Commission, 27 July 2026
- Timeline for the implementation of the EU AI Act, European Commission, AI Act Service Desk, checked 28 July 2026
- AI Act, Article 99: Penalties (amended text), European Commission, AI Act Service Desk, checked 28 July 2026
- Code of Practice on marking and labelling of AI-generated content, European Commission, checked 28 July 2026
- Artificial intelligence: Council gives final green light to simplify and streamline rules, Council of the EU, 29 June 2026
Facts and prices are as of the dates in the sources.
Nikolay Tolmachev
Founder, Tolman Studios
I build websites, AI chatbots and ads for small businesses, working from Slovakia. Before that I spent over 4 years managing IT projects. I hold PMP and Certified ScrumMaster certificates.