As of August, companies must disclose AI chatbots and label deepfakes. Does it apply to you?

What happened: As of August 2, 2026, new transparency rules under the EU's AI Act are in force. The Czech Telecommunication Office (ČTÚ), which oversees AI in Czechia, issued a press release around that date explaining who the rules apply to. They cover two groups: providers of AI systems (those who place them on the market) and deployers (those who use them in practice — typically a company running a chatbot on its website). Providers must make sure a person knows they're talking to an AI (chatbot, voice assistant, virtual avatar), and that AI-generated or AI-modified content (text, image, audio, video) is labeled in a machine-readable way. Deployers must additionally disclose any use of emotion recognition or biometric categorization, and label deepfakes intended for the public. Fines for violations can reach up to €15 million or 3% of a company's global turnover.

What it means for you: This isn't just about big AI companies, but the obligations are narrower than a quick news headline might suggest — it's not that every sentence or image an AI helped with now needs a visible tag. Specifically, two situations are covered:

• A customer-facing chatbot or voice assistant — whoever runs one should clearly state at the start of the conversation that it's AI, e.g. “you're talking to an AI assistant.” The duty technically falls on the provider of the system, but in practice it's the business deploying and customizing the chatbot for its customers that has to make it happen.
• Deepfakes and AI text for the public — if you publish image, audio or video content that creates or manipulates real people or events in a way that could be mistaken for genuine, you must clearly label it as artificially generated. The same goes for AI-generated text informing the public on matters of public interest, unless it's been reviewed and approved by a human with editorial responsibility.

An ordinary AI-assisted illustration for an article, or a rephrased paragraph that doesn't impersonate a real person or event, doesn't fall under this visible-labeling duty — the technical machine-readable marking of such content is primarily the AI tool maker's job, not the business using the tool.

A practical example: if your online store runs an AI-powered chatbot, adding one sentence to its opening message stating it's an AI assistant is a simple fix that saves you the risk of an inspection.

How to start:

1. Map out where in your business AI communicates directly with customers or the public — website, chatbot, phone line, social media.
2. Add a clear, visible AI disclosure at your chatbots and voice assistants.
3. If you publish deepfake content or AI-generated text aimed at informing the public on matters of public interest, label it as artificially created.
4. In unclear cases — an ordinary AI-assisted illustration, say, or internal tools with no public contact — check with a lawyer, since the scope of the obligation differs depending on whether you're a provider or just a deployer of someone else's tool.

What to watch out for: Makers of synthetic-content AI tools that were already on the market before August 2, 2026 get extra time under a recent amendment to implement the technical machine-readable marking duty — until December 2, 2026. But that applies primarily to AI tool providers, not to the visible disclosure a business runs as a deployer — no similar grace period applies there. Oversight in Czechia sits with the Czech Telecommunication Office, so this is a real inspection risk, not just a theoretical obligation.