Must a business label AI-generated content?

Not always. From 2 August 2026, under Article 50 of the AI Act, a business must make clear to people that they are dealing with a chatbot or another AI system unless this is obvious. It must also label deepfakes: realistic AI images, video or audio that could pass as a genuine record of real people, places or events. It must also label AI-generated text published to inform the public on matters of public interest, unless the text went through human editorial review. An ordinary advertising illustration that does not pretend to be real usually does not require an AI Act label. Breaches can be fined up to €15 million or 3% of turnover.

Chatbots and voice assistants: users must know they are talking to AI

Article 50(1) requires that people are told they are interacting with an AI system unless this is obvious from the circumstances. Technically the duty lies with the system's provider. A business that puts a chatbot on its website or customer line should still always show a notice, for example a line of text at the start of the conversation.

The information must be clear and given at the latest at the first interaction.

Deepfakes and AI text on matters of public interest

Anyone publishing an AI image, video or audio that resembles real people, objects, places or events and could appear authentic must disclose that the content was artificially generated or manipulated (Article 50(4)). Typical examples are a realistic video of a real person, an imitated voice, or a photo of an event that never happened.

For an evidently artistic, satirical or fictional work, a disclosure that does not spoil the work is enough.

AI-generated or AI-edited text that a business publishes to inform the public on matters of public interest must be labelled. The exception is text that went through human review or editorial control, with someone holding editorial responsibility for it.

What you need not label, and where to be careful

Company illustrations, icons or product visualisations that do not pretend to be a genuine record are usually not deepfakes. Text that AI helped to write, that a person checked and that the business takes responsibility for, does not have to be labelled.

Other rules may still require disclosure: rules against misleading advertising, social-network policies, and the protection of real people's personality and likeness. Using a specific person's face or voice without consent can be unlawful even with a label.

Provider duties, supervision and fines

Providers of generative AI must mark its output in a machine-readable way, for example with watermarks or metadata (Article 50(2)). Under Regulation (EU) 2026/1744, systems placed on the market before 2 August 2026 have until 2 December 2026. The European Commission has published a code of practice on labelling AI-generated content.

Supervision is national. In the Czech Republic it is the Czech Telecommunication Office. Breaches of the transparency duties can be fined up to €15 million or 3% of worldwide annual turnover.

A lawyer can help you decide what your business must label and how, and cover AI use in contracts with agencies and suppliers. Lawyers' services are paid; only submitting a standard request through LawBridge is free.

Frequently asked questions