AI Act Transparency: Must Your Chatbot Say It's AI?

From 2 August 2026 your chatbot must visibly disclose it is AI (AI Act Article 50). Not postponed. What counts as valid disclosure and what does not.

Gepubliceerd door NordX Consulting — AI bureau voor enterprise bedrijven in Nederland.

From 2 August 2026, your chatbot must tell people it is an AI. Not in the fine print, but visibly, in the conversation itself. This is Article 50 of the EU AI Act, and unlike the high-risk rules, this obligation was not postponed.

What Article 50 actually requires

Article 50 is about transparency: people must know when they are dealing with AI. It covers four situations.

1. Chatbots and virtual assistants. Any AI system that interacts directly with people, such as a customer service chatbot, a virtual assistant, or an automated phone system, must be designed so the user knows they are talking to AI. Exception: if it is obvious to a reasonably observant person that it is AI, it need not be stated explicitly.

2. Deepfakes. Anyone who generates or manipulates image, audio, or video content that constitutes a deepfake must disclose that the content has been artificially generated or manipulated.

3. AI-generated text on matters of public interest. Text created with AI and published to inform the public on matters of public interest must be marked as artificial.

4. Emotion recognition and biometric categorisation. Anyone deploying such systems must inform the individuals exposed to them.

For most Dutch businesses, point 1 is the most relevant: if you run a chatbot on your website or in customer service, it falls under this.

What counts as valid disclosure, and what does not

This is where it goes wrong. The disclosure must be perceivable in the interaction itself. Concretely:

Sufficient:

A visible line at the start of the conversation: "You are chatting with an AI assistant from [company]."
A clear label on the chat bubble or window indicating the AI nature.

Not sufficient:

A sentence hidden in the terms and conditions or privacy policy.
A metadata watermark alone that the user never sees.
A vague label like "assistant" or "helpdesk" that does not make clear it is AI.

The rule is simple: the user must be able to see it at the moment they interact with the system, not somewhere deep in a document no one reads.

What was and was not postponed

The June 2026 news was "the AI Act has been delayed." That is partly true, and the part that is not true is exactly this.

The high-risk obligations moved to December 2027 (see what was and was not deferred). Article 50's transparency duty did not: it applies from 2 August 2026. There is a proposal to give the marking obligation for AI-generated content specifically (Article 50(2)) targeted time until 2 December 2026, but the core, the disclosure duty for chatbots, stands as of 2 August.

The penalties

Non-compliance with Article 50 can reach 15 million euro or 3 percent of global annual turnover, whichever is higher. That is the ceiling, not the standard penalty. But it signals how seriously the legislator takes transparency.

What to do now

1. Inventory your AI touchpoints. Chatbots, virtual assistants, automated phone or email replies, AI-generated images on your site. Make a list.

2. Add visible disclosures. For every chatbot: a clear line that the user is talking to AI, visible at the start of the conversation. This is usually a copy change, not a rebuild.

3. Label AI-generated media. Using AI for images or video aimed at the public? Add a visible label.

4. Record it. Document which systems use AI and how you handled the transparency. That makes any inspection far simpler.

Most organisations finish this within a day. It is not a high-risk project with documentation and risk assessment; it is a matter of being honest about where AI is involved, in a place the user can see.

In short

As of 2 August 2026, your chatbot must visibly disclose that it is AI. Hiding it in the terms does not count. The same duty applies to deepfakes and AI content on matters of public interest. The obligation is not postponed, the penalty is substantial, and the fix is usually a few lines of adjusted text.

The transparency duty is one part of the AI Act. See the full overview in our AI Act compliance guide for Dutch businesses.

Sources

Frequently asked questions

Does my customer service chatbot have to disclose it is AI?

Yes. Article 50 of the EU AI Act requires, from 2 August 2026, that users know they are interacting with an AI system. The disclosure must be visible in the conversation itself, not hidden in the terms and conditions.

Was the transparency obligation postponed like the high-risk rules?

No. The high-risk obligations moved to December 2027, but Article 50's transparency duty applies from 2 August 2026. Only the marking obligation for AI-generated content (paragraph 2) may receive targeted time until 2 December 2026.

What counts as valid AI disclosure?

A disclosure the user perceives during the interaction, for example a visible line at the start of the chat. A sentence in the terms and conditions or an invisible metadata watermark alone does not qualify.

How large are the penalties for non-compliance with Article 50?

Non-compliance can reach 15 million euro or 3 percent of global annual turnover, whichever is higher.

Meer weten? Bekijk onze andere artikelen op het NordX blog of neem contact op via nordx.ai.