From August 2, the AI transparency duty bites – what SMEs should do now

From August 2, 2026, Article 50 of the EU AI Act applies. Anyone communicating with customers via an AI chat or publishing AI-generated text, images and voices must disclose it. For most SMEs this is no reason to panic – but a reason to settle three things this week. I write this as someone who runs several shops himself and had to answer the same questions for his own businesses.

What it's really about

The public debate revolves around fines and doomsday. In practice the message is soberer: transparency. Lawmakers want people to know when they're talking to a machine and when content was machine-generated. That's not a technical hurdle but a question of honesty toward the customer – and most reputable businesses already meet the spirit of the rule, they just haven't documented it.

What actually applies

Two duties touch almost every business with an online presence:

• A chatbot or AI phone assistant must identify itself as AI – not in the fine print but where the conversation begins.
• Synthetically generated content – product copy, images, audio, deepfakes – must be labelled as such and machine-readable.

In parallel, the Commission's enforcement powers over general-purpose AI model providers take effect. The high-risk deadlines postponed in spring (now late 2027) explicitly don't change these two dates – a point many compliance discussions got confused about recently.

What it means for your shop

In practice a clear notice in three places is usually enough: in the chat window ("You're chatting with an AI assistant"), under AI-assisted product copy, and on automatically generated images. What matters isn't perfect legal wording but that the labelling exists, is consistent and is traceably documented – that's what authorities check first. A business that can show where and how it labels is in far better shape than one with a perfect clause and chaotic execution.

The checklist for this week

• Inventory: Where does an AI talk to our customers – chat, email assistance, phone bot, social replies?
• Content: Which published texts and images are AI-generated and not yet labelled?
• Ownership: Who is named internally if a regulator asks – and where is the documentation?
• Vendors: Do we use tools that themselves fall under provider duties, and do we have their disclosures on file?

My advice, entrepreneur to entrepreneur

Anyone who can answer these four is further along than most. The rest is implementation – and that's manageable. Don't treat the deadline as a fear topic but as a reason to write down, once and cleanly, where AI touches your customers. You need that map anyway – for the regulation, but above all for yourselves.