TL;DR
- Since August 2, 2026, the EU AI Act requires clear labeling of AI-generated audio, images, video, and text
- The label must be in a machine-readable format so it can be detected automatically
- Deepfakes need a visible disclosure that the content was generated or manipulated by AI
- AI text published to inform the public must be disclosed, unless a human reviewed and owns editorial responsibility for it
- Fines for breaking these transparency rules go up to 15 million euros or 3% of global turnover, with lower caps for small companies
You have probably used ChatGPT to write a product description or Gemini to generate an image for a social post. That is fine. Starting this week, how you publish those outputs changed.
The EU AI Act's transparency rules started applying on August 2, 2026. If your business deals with customers in the European Union, exports content there, or runs ads across EU platforms, this affects you. Even companies outside the EU that target EU audiences are in scope.
Here is what the law actually says, stripped of legal jargon, and what it means for a small business.
What the new labeling rule requires
The core obligation is simple: content created or heavily edited by AI must be marked in a way that shows it was machine-made.
The law uses the term "artificially generated or manipulated." That covers synthetic audio, images, video, and text. The key word is synthetic. A photo you took and slightly brightened in an editor is still your photo. A face-swapped video is not, and it must be flagged.
The label is not only a visual note. It has to be machine-readable, meaning software can detect it without human help. Think of it as a digital watermark or embedded metadata that survives when the file moves around.
The rules apply to providers (the companies building the AI tools) and deployers (the businesses that use those tools to publish content). If you run the AI, you carry part of the responsibility for what comes out of it.
The practical rules that matter
You do not need a lawyer to follow this. The cases that come up most often look like this.
Deepfakes need a visible disclosure. If you generate or manipulate a video, image, or audio clip that looks real but is not, you must tell viewers it is fake. This includes face swaps, cloned voices, and altered footage. The disclosure must be clear and easy to notice, at the time of first exposure.
Public-interest text needs a label too. This is the one most content teams miss. AI-generated text published to inform the public must be marked as AI-generated. A news roundup, an industry explainer, a market update, those count. There is an escape hatch: if a human reviewed the text and someone takes editorial responsibility for it, the disclosure is not required.
Artistic and satirical work gets a lighter touch. Creative, fictional, or satirical content still needs disclosure, but it does not have to break the viewer's experience. A subtle credit can be enough.
Where the fines come in
The EU AI Act is not a suggestion. Violating the transparency obligations under Article 50 can cost up to 15 million euros or 3% of total worldwide annual turnover, whichever is higher.
Small businesses get relief. For SMEs and startups, fines are capped at the lower of those two numbers, and authorities must consider the economic viability of the company. The message is clear: this is about deterrence, not bankrupting a bakery that posted an AI-made ad.
What Moldovan businesses should actually do
Moldova is not in the EU yet, but it is a candidate country and its laws are aligning with EU rules as part of the accession talks. The law follows the audience, not the headquarters. If you sell to EU customers, advertise on EU platforms, or publish content aimed at EU readers, the transparency rules can reach you.
A few cheap habits keep you safe.
Disclose AI images and videos. Add a short note under AI-generated visuals: "This image was created with AI." For ads and promotional video, make the disclosure visible, not buried in a caption in tiny print.
Flag AI text that informs the public. If you publish a blog post or newsletter generated by AI, either label it clearly or have a person review it and own it. Human review plus editorial responsibility removes the obligation. That is a realistic path for most businesses.
Keep the machine-readable mark when you can. Platforms are already building tools to add these marks automatically. When your AI tool offers embedding a label, turn it on. It costs nothing and removes the risk.
"The transparency rules are built around one idea: people should know when a machine made what they are seeing."
What this means for the next twelve months
Expect the ecosystem to adapt fast. Major AI platforms will bake labels into their outputs by default, and detection tools will get better at finding content that slipped through. Platforms that host user content will start asking about AI involvement in uploads.
For business owners, the practical answer is boring and good: keep a human in the loop, label what AI made, and save the receipts. That approach costs almost nothing and keeps you out of fine territory. It also builds trust, which is harder to buy than any tool.
Start with your most visible AI content today. It takes five minutes to add a disclosure line, and it keeps you ahead of both the law and the competition.
ShiftIT is a custom software and AI automation agency based in Chișinău, Moldova. We help businesses automate operations and scale their digital presence.