Ilmari Salmisto
If you use AI in your business, the EU AI Act already applies to you. The good news is that the heaviest obligations were pushed back to 2027. The bad news is that the transparency rules come into force on 2 Aug 2026, and they apply to any ordinary business that runs a chatbot, makes AI images, or generates text with AI. Don't panic. This is easier than it sounds.
I've talked with plenty of business owners who feel the same way they did during GDPR. A big legal package arrives from Brussels, the headlines shout about 35 million euro fines, and a nagging voice in the back of your head says you probably need to hire a lawyer now. Most of the time you don't. Let's go through what actually changes on 2 Aug 2026, who it applies to, and how to get it sorted without the panic.
The short version, if you're in a hurry:
- The heavy obligations for high-risk AI moved from 2 Aug 2026 to 2 Dec 2027. They don't apply to most businesses.
- The transparency rules (Art 50) come into force on 2 Aug 2026. They apply to any ordinary business using AI.
- In practice: tell the user when they're chatting with a bot, and label any image, audio, or video made with AI.
- For an ordinary business, a transparency breach is capped at 15M euros or 3% of turnover, not the 35M euros from the headlines. SMEs are subject to the lower cap.
- You can get this sorted in a few hours, not months.
What is the EU AI Act?
The EU AI Act (formally the AI Regulation) is the shared EU-wide rulebook for how AI may be developed and used. It entered into force on 1 Aug 2024 and applies in stages over several years. The Act classifies AI by risk: prohibited practices, high-risk systems, and the more lightly regulated everyday AI that most businesses fall under.
The idea is simple. The bigger the risk to people, the stricter the rules. Facial recognition in a public space is a different matter from a chatbot in an online store. Most SMEs use AI for ordinary things: customer service, content, reporting. For them it mostly comes down to one thing, transparency. If you want background on how to bring AI into a business in the first place, start with AI for business.
What changes on 2 Aug 2026, and what doesn't
The most important date for an ordinary business is 2 Aug 2026. That's when the AI Regulation's transparency obligations (Article 50) come into force, and the enforcement and penalty machinery switches on. The heavy obligations for high-risk systems, on the other hand, were postponed by the so-called Digital Omnibus amendment, which the European Parliament confirmed on 16 Jun 2026.
This distinction is what the headlines often hide. The delay applies to high-risk AI, not to transparency. Here's what moved and what didn't:
| Obligation | Who it applies to | In force |
|---|---|---|
| Transparency (Art 50): disclose the bot, label AI content | Any ordinary business using AI | 2 Aug 2026 (not delayed) |
| Machine-readable marking of synthetic content | Providers of AI models | 2 Dec 2026 |
| High-risk AI, Annex III (e.g. recruitment, credit) | Deployers of high-risk systems | 2 Dec 2027 (delayed from 2 Aug 2026) |
| High-risk AI, Annex I (regulated products) | Product manufacturers | 2 Aug 2028 (delayed) |
So if you don't do recruitment, credit scoring, or similar high-risk decisions with AI, the heavy obligations won't touch you for a long time yet. But transparency touches you right away if you use bots or AI-made content.
Who do the transparency rules actually apply to?
The transparency rules apply to you if your business uses AI in a way that isn't obvious to the user. In practice, three situations cover most businesses. You don't need a lawyer to spot them, just go through your own operations.
- A chatbot or AI customer service. If a customer is talking with AI, they have to be told. Clearly and immediately, on the first message, not hidden at the bottom of the page.
- Images, audio, or video made with AI. If you publish a deepfake or other media produced or edited with AI, it has to be labelled as AI-generated.
- AI-generated text published to the public. AI text published for public information has to be labelled, unless a human has reviewed it editorially and takes responsibility for it.
That last point matters especially for marketing agencies and content creators. If AI writes a draft and a human edits it and stands behind the content, no separate label is needed. Responsibility for what you publish stays with a human. The same goes for biometric identification and emotion recognition: if you use them, you have to disclose it, and GDPR still applies on top.
Transparency isn't an obstacle. It's the same thing as being honest with the customer. Most good businesses already do this, the law just writes it down.
How big are the fines, really?
For an ordinary business, the cap on a transparency breach is at most 15 million euros or 3% of global turnover, whichever is higher. The 35 million from the headlines applies to prohibited practices, such as manipulative AI or unauthorised biometric surveillance, things an ordinary business doesn't do. In practice SMEs are subject to the lower cap.
| Type of breach | Fine cap |
|---|---|
| Prohibited practices (Art 5) | 35M euros or 7% of turnover |
| Other obligations, incl. transparency (Art 50) | 15M euros or 3% of turnover |
| Providing incorrect information to authorities | 7.5M euros or 1% of turnover |
In practice, nobody is going to hand an SME a 15 million euro fine because a bot forgot its disclosure. The numbers are caps, and enforcement looks first at severity and intent. But you should still get this sorted, because it's easy and because your customers' trust is worth it.
How to sort it out with minimal effort
An ordinary business can get the AI Regulation's transparency sorted in a few hours. You don't need a systems project or an ongoing legal bill. Go through these points once, fix what's missing, and note down what you did. Here's a concrete checklist:
- List where you use AI. Bots, image generators, text generators, audio. You can't label what you don't know is there.
- Add a clear disclosure to your bots. One sentence: "This is an AI assistant." Right on the first message.
- Label AI-generated media. Images, videos, and audio made or edited with AI.
- Agree on text review. Decide who reviews AI text and takes responsibility for it. Then no separate label is needed.
- Note down what you did. A short memo is enough. If someone asks, you have an answer ready.
If you build AI solutions with us, transparency comes built in. When we build a bot, it says it's a bot, and when we build a content automation, we plan the labelling in advance. It's part of a good build, not a separate add-on. You can read how PolkuAI builds a solution from discovery to maintenance, or go straight to services.
Frequently asked questions
Does the EU AI Act apply to a small business?
Yes, if you use AI. Company size doesn't exempt you from the obligations, but for most SMEs it's only about transparency: tell people when a customer is chatting with a bot, and label any image, audio, or video made with AI. The fine caps are large, but SMEs are subject to the lower cap and enforcement looks at the severity of the breach.
What do you need to do before 2 Aug 2026?
Make your use of AI transparent. Add a disclosure to your bots that they are AI, label AI-generated media, and agree on who reviews AI text and takes responsibility for it. For an ordinary business this is a few hours of work, not a months-long project.
Were the EU AI Act obligations delayed?
Some were. The heavy obligations for high-risk systems were postponed by the Digital Omnibus amendment: Annex III systems to 2 Dec 2027 and Annex I products to 2 Aug 2028. The transparency obligations were not delayed, they come into force on 2 Aug 2026 as planned.
Can you publish AI-generated text without a label?
Yes, if a human has reviewed the text editorially and takes responsibility for it. This matters for content creators and marketing agencies. Text produced purely by AI and published to the public as-is has to be labelled, but edited content that a human stands behind doesn't require a separate label.
Where can I get help with AI compliance?
You can handle transparency yourself with this checklist, it's usually a few hours of work. If you build AI solutions, it's worth making compliance part of the build from the very start. PolkuAI builds solutions so that transparency is included from the outset. You can book a free discovery call where we go through your situation together.