EU explains when AI must identify itself as AI
July 21, 2026
The European Commission has published guidance on Article 50 of the AI Act. For chatbots, deepfakes, and AI agents, practical transparency work now starts before August 2, 2026.
What this is about
The European Commission published guidelines on July 20, 2026, for the transparency obligations under Article 50 of the AI Act. These obligations apply from August 2, 2026. That turns an abstract legal rule into a concrete product question: when must a system tell people they are interacting with AI?
The timing is tight. Providers and deployers have only days to compare interfaces, labels, and internal evidence against the guidance.
What the guidelines actually do
The guidelines explain how competent authorities, providers, and deployers should interpret the transparency rules. They cover direct interaction with people, machine-readable marking of synthetic content, deepfake disclosures, and certain AI-generated public-interest text.
The reading for AI agents is especially important. William Fry notes that an agent should disclose not only its artificial nature but also on whose behalf it is acting. That matters once a system does more than answer questions: booking appointments, sending messages, or triggering processes.
Why it matters
Transparency is not decoration. People make different decisions when they know whether they are dealing with a person, a bot, or an authorized agent. That affects customer service, hiring, political communication, education, health, and support.
For companies, the obligation is also an architecture issue. A notice at the end will not always be enough. If AI is embedded across channels, organizations need consistent labelling, logs, ownership, and an answer to what happens when an agent acts on behalf of a person or company.
The EU pairs these guidelines with a voluntary Code of Practice for AI-generated content. Using it can provide more legal predictability, but it does not remove responsibility for implementation.
In plain language
Imagine a delivery service. It matters whether a friend calls for you, a call center calls, or an automated system places the order. The guidelines ask for a clear name badge: who is speaking, is it AI, and who is it acting for?
A practical example
An insurer lets an AI agent triage claims. Each day, the agent handles 8,000 chats, asks for photos, and suggests next steps. From August 2, 2026, it is not enough to mention “automated support” somewhere in the terms.
The interface must clearly show that an AI system is interacting. If the agent acts on behalf of the insurer, that must also be clear. For deepfake or synthetic media in the process, the company needs suitable labels and traceable internal rules.
Scope and limits
First, the guidelines are not a complete technical specification. They describe what must be achieved, but not the exact component required for every interface.
Second, transparency does not solve correctness. A properly labelled chatbot can still give wrong, unfair, or dangerous recommendations.
Third, implementation becomes difficult for multi-step agents. If an agent connects multiple tools, subagents, or human approvals, the disclosure still needs to remain understandable.
SEO & GEO keywords
EU AI Act, Article 50, AI transparency, AI Agents, Deepfake Labelling, European Commission, AI Office, Synthetic Content, Compliance, AI Regulation
💡 In plain English
The EU is clarifying when people must be told they are talking to AI or seeing AI-generated content. For companies, this is not just a notice, but a design and evidence obligation.
Key Takeaways
- →The Article 50 guidelines were published on July 20, 2026.
- →The transparency obligations apply from August 2, 2026.
- →AI agents should disclose that they are AI and on whose behalf they act.
- →Labelling does not replace checks for correctness, fairness, or safety.
FAQ
When do the transparency obligations apply?
The Article 50 obligations under the AI Act apply from August 2, 2026.
Is a notice in the terms enough?
Often no. The information must be understandable to the affected person and appropriate to the interaction.
Does this also affect AI agents?
Yes. The guidelines read direct AI interaction broadly and include agents when they interact with natural persons.