One click, and your vacation photo becomes an ad in which you endorse a product you've never seen. The EU wants to curb cases like that from August 2. The price: from then on the AI Act's transparency rules kick in — and they don't just concern Big Tech, they potentially concern you too.
What changes on August 2
The AI Act is the EU's AI law. From August 2, 2026, its transparency obligations (Article 50) apply. Simplified, three things:
- Providers of AI systems (OpenAI, Google, Meta …) must mark generated content in a machine-readable way — an invisible signal that lets software tell: "this came from an AI."
- Anyone who publishes AI content and builds a deepfake in the process (an image, video or audio that shows real people convincingly) must disclose it visibly.
- Chatbots must say you're writing to a machine, not a human.
Sounds reasonable. The devil, as always, is in the details.
What you as a creator need to watch
Short version: less than the panic headlines suggest.
- An AI image of a real person that could pass as real? Label it. A "made with AI" underneath is enough.
- Your own AI chatbot on your site? A note "you're chatting with an AI" does the job.
- AI texts on matters of public interest? Disclose — unless a human reviews them and holds editorial responsibility.
Not affected: abstract or obviously artistic AI images (no deepfake), purely private use, and anything that already says "made with AI." So for most hobby creators … not much changes.
And now the criticism
Still, the rule has a problem — a big one. It's vague where it should be precise, and strict where a sense of proportion would have done.
"Public interest," "convincingly real," "editorial responsibility" — rubber terms that, in case of doubt, only a court will settle. For a corporation with a legal department, that's an annoyance. For a small team, a start-up, a single creator, it's an intimidation factor: better to do nothing than risk a fine — up to 15 million euros or 3% of annual turnover.
That's exactly where the competitive disadvantage lies. While the US and China just build and ship — see Meta's Muse Image or Google's Gemini images, both "not in the EU for now" — Europe debates labeling symbols. The effect is absurd: the rule is meant to protect European users, yet it ensures they don't get the tools in the first place. Protection by absence is no protection.
And then the objection the whole lovely idea shatters on. Picture the fraudster who dutifully tags his phishing deepfake with "🤖 AI-generated" before he strikes. You can practically see him polishing the disclaimer. Hard to imagine? Exactly. Anyone building a deepfake with criminal intent — for phishing, fraud, character assassination — will of course never label it. Criminals, as a rule, aren't the crowd that sticks meticulously to regulations. So the obligation reliably hits the honest, while its actual target audience carries on, relaxed — fine or no fine.
To be fair: it's not entirely pointless. In the gray zone — satire, advertising, the clueless resharing of a fake video — a note creates real clarity. But against the hard abuse that makes the headlines, no watermark helps — only law enforcement does. And that would exist without a new AI law too. When a rule mainly paralyzes the small players and waves the big ones through with lawyers, while the actual culprits stay untouched, it misses its purpose.
What this means for this blog
The honest self-test, since we demand it of others too: does the AI Act concern us?
Our header images come from an AI (Stable Diffusion) — but abstract, no real people, so no deepfakes. Our texts are written with AI — but under every article, from day one, it says "✍️ Written with …". We've long labeled voluntarily what the law demands.
Result: we don't need to change a thing. Not because we're weaseling out, but because transparency was part of the idea anyway. Maybe that's the real lesson — labeling costs almost nothing if you build it in from the start. The expensive part is the uncertainty about whether you're doing it right.
And that one, Brussels brought on itself this time.