The EU AI Act: what changes from August 2, 2026 and why it concerns your business

If you're hearing about the "EU AI Act" for the first time - you're not alone. It's rarely explained in plain language: lawyers write about it in a complicated and intimidating way, and businesses find out about the new rules after the fact.

I'll explain it calmly and step by step: what this law is, who it concerns, what exactly changes on August 2, and - separately - how it affects paid advertising. Without the legal fog.

What this law even is

The AI Act (full name - Regulation (EU) 2024/1689) is the world's first comprehensive law regulating the use of artificial intelligence. It was adopted in the EU and is officially published by the European Commission. The idea is simple: the higher the risk from using AI, the stricter the rules. The law divides all AI systems into several levels - from "minimal risk" (almost no requirements here) to "high risk" and outright prohibited practices.

It's being introduced not all at once, but in stages, from 2025 to 2028. And August 2, 2026 is one of the key dates.

Who it concerns (spoiler: almost everyone)

Here's the main misconception that makes many people relax. The law concerns not only IT companies and not only those who "make AI." It concerns any business that uses AI tools - and today that's practically everyone: generating texts and images with neural networks, chatbots on a website, AI tools for hiring, advertising automation.

And two more important points. First, the law has no exemptions for small businesses - company size doesn't matter. Second, it applies not only within the EU: if the output of your AI reaches users in the European Union, you fall under the law, wherever your company is registered. You work from Prague, Kyiv, or anywhere else, but your ad, bot, or content is seen by Europeans - the rules are yours.

What exactly changes on August 2, 2026

The main thing that takes effect on this date is the transparency rules (that's Article 50 of the law). To put it very simply, the essence is one thing: a person must understand when they're dealing with artificial intelligence rather than a real human or real content. Let's break it down point by point.

Chatbots must admit they're bots. If you have an AI bot working on your website, in a messenger, or in an ad, the user must clearly understand they're talking to a machine. And you have to notify them right away, at the moment the conversation begins - not hide it in small print in the terms of use.

AI content must be marked. Texts, images, audio, and video created by a neural network must have a technical mark indicating they're artificially generated (in a machine-readable format, so platforms can recognize it).

Deepfakes must be labeled explicitly. If you've used AI to create content that looks real - a real person, a real event - and it could mislead, you're obligated to mark it as artificial.

People must be warned about emotion recognition and biometrics. If a system tries to determine a person's emotions or categorize them by biometrics, people must be informed of this.

There are exceptions to the rules - and this is important. You don't need to mark things where the artificiality is already obvious, as well as artistic, satirical, or clearly fictional content. And separately: if AI content has undergone real human editing and someone has taken responsibility for it, the requirements are eased. So there won't be total labeling of "everything under the sun" - it's about what could mislead someone.

An important clarification so you don't get confused

How this affects paid advertising

And now - the most interesting part for you and me. Advertising is increasingly made with the help of AI: generative creatives, AI voiceovers, virtual avatars, automatic texts, bots for receiving inquiries. The law directly affects these tools. Here's what's important in practice.

AI creatives and deepfakes. If you generate advertising images or videos with a neural network - that's allowed in itself. But if the creative features a "seemingly real" person, event, or, all the more so, a recognizable face created by AI - such content must be marked as artificial. Be especially careful with synthetic "people," talking avatars, and voices: this is the deepfake zone.

Automation in ad accounts. Tools like creative generation inside Meta or Google formally mark content on their side, but responsibility for the honesty of the advertising still lies with you as the advertiser. It's worth understanding what exactly AI generates in your campaign, and not passing off the generated material as real footage if it could mislead the audience.

Bots in advertising. Running ads that lead into a conversation - Click-to-Messenger, WhatsApp, a bot for qualifying leads? The bot must honestly indicate that it's a bot. This doesn't get in the way of sales, but it must be there from the first message.

Targeting - you can exhale. The law doesn't prohibit ordinary targeting by interests and behavior. Ad personalization is normal and legal. The line is drawn where manipulation begins: hidden influence that bypasses a person's conscious choice, or exploitation of vulnerabilities (more on this below).

Your team must understand AI. The law introduces an "AI literacy" requirement: people who use AI tools in their work (including in advertising) must understand at a basic level how these tools work and where their limits are. This isn't an exam, but it's not an empty formality either.

What's already prohibited (and in effect since 2025)

Separately, it's worth knowing about prohibited practices - they took effect even earlier, in February 2025. For marketing, this means: you can't use AI for hidden manipulative techniques that distort a person's behavior and cause them harm, and you can't exploit people's vulnerabilities - by age, disability, or difficult financial situation. Simply put, aggressive AI-based "dark patterns" that pressure the most vulnerable are no longer a gray area, but a direct violation.

How much it costs to violate

The fines are serious, and this isn't to intimidate but so you understand the scale. For violating the transparency rules - up to 15 million euros or 3% of the company's annual worldwide turnover (the larger amount is taken). For using prohibited practices - up to 35 million euros or 7% of turnover. It's clear that the maximums are reached in extreme cases, but the order of magnitude itself shows: this isn't a topic to ignore.

What a business should do - a short plan

Nothing heroic is required, but it's worth going through this list before August.

• Make a list of the AI tools you use - in advertising, on your website, in customer service. Just to understand where you even have AI.
• Check your chatbots: is it clear everywhere that it's a bot, and is this communicated right away?
• Mark AI content that could mislead - especially deepfakes and synthetic "people" in advertising.
• Arrange for human editing: let a specific person be responsible for the AI content that goes to publication. This both eases the requirement and protects quality.
• Run a short briefing for the team - so that those who work with AI understand the rules of the game.

The key points in brief

On August 2, 2026, transparency rules for AI take effect in the EU. They concern any business that uses AI - regardless of size and of where it's registered, if its audience is in the EU. The essence is simple: a person must understand when AI is in front of them. Chatbots indicate they're bots; generated and misleading content is marked; manipulative practices have been prohibited since 2025. For advertising, this means being more attentive to AI creatives, deepfakes, and bots - and keeping a human responsible for what goes on air. Nothing scary, if you figure it out in advance.