Agent Disclosure Rules

What are Agent Disclosure Rules?

Agent disclosure rules refer to the transparency obligations established under Article 50(1) of the EU AI Act. These rules mandate that providers of AI systems intended to interact directly with natural persons must inform those users that they are interacting with an AI, rather than a human.

How the Rules Work

The obligation to disclose is triggered when four cumulative criteria are met:

1. AI system definition: The technology must qualify as an AI system under the EU AI Act definition, explicitly excluding non-AI automation such as rule-based quick answers.
2. Intended for natural persons: The system must be intended to interact with natural persons via auditory, visual, or physical form.
3. Direct interaction: The interaction must be direct, meaning it occurs in real-time or near real-time—not a backend machine-to-machine call.
4. Natural person: The interaction must be with a natural person, not another machine.

This scope is broad, encompassing AI agents, chatbots, voice assistants, digital avatars, humanoid robots, and companion devices. The European Commission’s draft Guidelines on transparency obligations clarify that AI agents must be designed to disclose their AI nature in every human interaction, and must identify the principal on whose behalf they act.

A Practical Example

To understand how these rules apply, it helps to look at how the four criteria interact across different types of systems. The obligation is not a blanket requirement for all AI, but a conditional one based on the nature of the interaction.

Consider these three scenarios:

• The Customer Service Chatbot: This system is designed for direct, real-time interaction with a human user. It meets all four criteria: it is an AI system, intended for natural persons, the interaction is direct, and the user is a natural person. Consequently, the provider must ensure the user is aware they are speaking to an AI. A notice such as “You are interacting with an AI system” before substantive interaction satisfies the requirement. Burying this in terms of service or help pages is insufficient.
• The Autonomous Coding Agent: This agent primarily interacts with code repositories and build systems. However, it may occasionally interact with developers via chat interfaces to report bugs or request clarification. Here, the “direct interaction” criterion becomes ambiguous. If the agent is designed to engage with developers in real-time, it likely triggers the disclosure requirement during those specific human-facing sessions, even if its primary function is machine-to-machine.
• The Backend Workflow Agent: This agent operates entirely in the background, orchestrating data between databases and APIs without any human intervention. Because there is no direct interaction with a natural person, the disclosure obligation does not apply. The system fails the “intended for natural persons” and “direct interaction” criteria, placing it outside the scope of Article 50(1).

The only notable exception is when the AI nature of the system is already obvious to a reasonably well-informed, observant, and circumspect person within the specific context of the interaction. However, the Commission Guidelines suggest this is a high bar—if there is any doubt, disclosure is required.

The Code of Practice Gap

The industry Code of Practice on Transparency of AI-generated Content covers synthetic content marking (Article 50(2)) and deepfake disclosure (Article 50(4)), but it explicitly excludes Article 50(1) agent disclosure. This means companies must navigate compliance by interpreting the regulation text and Commission Guidelines directly, rather than relying on a standardized industry framework for agent-specific disclosure.

This gap is particularly significant for AI agent developers, who must build disclosure mechanisms into their interaction flows. While the Commission’s Guidelines provide interpretive guidance, they are non-binding; authoritative interpretation may ultimately rest with EU courts as enforcement matures.

Penalties and Enforcement

Compliance is not optional. Non-compliance can result in penalties reaching up to €15 million or 3% of a company’s total worldwide annual turnover, whichever is higher. These rules have extraterritorial reach: providers based outside the EU must comply if their AI systems are placed on the EU market or if their outputs are used within the EU.

Important distinctions:

• Open-source not exempt: AI systems released under free and open-source licenses are explicitly included in Article 50 obligations under Article 2(12).
• No grace period: Unlike Article 50(2) machine-readable marking, which benefits from a transitional period for systems already on the market, Article 50(1) agent disclosure has no grace period.
• Enforcement date: Article 50 applies from August 2, 2026.

Common Questions

Does this apply to my open-source project?

Yes. Open-source AI systems are explicitly included in the scope of Article 50 obligations. The general open-source exemption in Article 2(12) does not apply to transparency obligations. If your open-source AI system interacts directly with humans, you must ensure disclosure requirements are met.

What counts as “obvious” AI?

The exception applies when the AI nature is obvious to a reasonably well-informed, observant, and circumspect person in context. The Commission Guidelines suggest this is a high bar—if there is any doubt, disclosure is required. Providers that cannot reliably predict whether an agent will interact with a human should design it to disclose in every interaction.

How does this relate to agent compliance?

Agent disclosure rules are one component of the broader agent compliance landscape. While agent compliance covers risk classification, human oversight, audit trails, and incident reporting across multiple jurisdictions, agent disclosure rules specifically address the transparency obligation when AI agents interact with humans under EU law.

Why It Matters

Understanding these rules is essential for several reasons. First, the enforcement date has passed, making compliance immediate for developers integrating necessary disclosure flows. Second, because the Code of Practice on Transparency of AI-generated Content explicitly excludes Article 50(1) agent disclosure, companies cannot rely on industry self-regulation to navigate these requirements; they must engage directly with the regulation and official guidelines. Third, the extraterritorial reach of the law means that developers based outside the EU are not exempt if their systems are used within the European market. Finally, the explicit inclusion of open-source AI systems under Article 2(12) closes the common assumption that open-source projects are exempt from these transparency obligations.

Sources

[1] European Parliament and Council, “Regulation (EU) 2024/1689 (EU AI Act)” (June 13, 2024).

[2] EU AI Act Explorer, “Article 50: Transparency obligations” (2024).

[3] European Commission, “Guidelines on transparency obligations for providers and deployers of certain AI systems” (July 20, 2026).

[4] Forkast, “Europe published agent disclosure rules; the industry code that covers everything else doesn’t mention them” (August 7, 2026).

[5] European Commission, “Code of Practice on Transparency of AI-generated Content” (2026).