EU AI Act: transparency rules for chatbots and AI content now apply

On 2 August 2026, a new part of the EU AI Act started to apply: the transparency rules. They are not about how an AI system is built, but about what people must be told when they meet one. For companies that put chatbots, voice assistants or AI agents in front of customers or employees, this is now a legal requirement, not only good practice.

This post is a summary for integration and platform teams, not legal advice.

What now applies

According to the law firm Cooley, three of the four transparency obligations became enforceable on 2 August:

• Interactive AI systems. Providers must make sure people know they are talking to an AI system, such as a chatbot or voice assistant, unless it is already obvious.
• Biometric systems. Deployers must inform people when emotion recognition or biometric categorisation is used.
• Deepfakes and AI-generated text. Deployers must disclose content that is artificially generated or manipulated, such as text on matters of public interest, unless a person has reviewed and edited it.

The fourth obligation, that providers of generative AI must mark synthetic audio, images, video and text in a machine-readable way, has a transition period for systems already on the market, until 2 December 2026.

Fines for breaking the rules can reach 15 million euros or 3% of global annual turnover, whichever is higher. The rules apply to AI placed on the EU market, wherever the provider is based.

Why it matters for integration teams

Transparency sounds like a job for the product owner or the legal team, but much of the work lands in integration. An agent that answers a customer by email, updates a ticket or writes a message in Slack often does so through an integration platform. The label that says “this was written by AI” has to be added somewhere along that path, and it has to survive when the message passes through several systems.

The same goes for logging. To show that the rules are followed, a company needs to know which messages and documents an AI system produced and which were reviewed by a person. That is information the integration layer and the AI gateway are well placed to record.

What to check

1. Where do agents talk to people? List the chatbots, voice assistants and agents that reach customers or employees, including those built inside integration platforms.
2. Does the AI label survive? Check that the notice that content is AI-generated is not lost when a message is transformed or passed on between systems.
3. Can you show human review? Where you rely on the exception for content reviewed by a person, make sure the review is logged.