EU AI Act Article 50(1): Chatbot Disclosure Requirements — Complete Compliance Guide

If your AI system talks to people, those people must know they are talking to AI. Article 50(1) of the EU AI Act makes this a legal requirement from August 2, 2026 — with fines up to €15 million for non-compliance. This guide covers exactly what you must do, how to do it, and what happens if you don’t.

1. What Article 50(1) Requires

Article 50(1) of Regulation (EU) 2024/1689 — the EU AI Act — establishes a single, clear obligation for AI systems that interact directly with people:

“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and context of use.” — EU AI Act, Article 50(1)

The obligation has three components. First, the duty falls on providers — the companies that build or place AI systems on the market. Second, the disclosure must happen at or before the first interaction, not buried in documentation that users will never read. Third, there is a narrow exception for cases where the AI nature is already obvious.

This obligation applies from August 2, 2026. There is no transitional period, no grace window, and no phased rollout. On that date, every AI system that directly interacts with people in the EU must either disclose its AI nature or qualify for the “obvious” exception.

2. Who Must Comply: Providers vs Deployers

Article 50(1) places the obligation on providers — the entities that develop an AI system or have it developed and place it on the market or put it into service under their own name or trademark. If you build a chatbot and offer it to businesses or consumers, you are a provider and the disclosure mechanism must be built into your system.

Deployers — companies that use a third-party AI system — are not directly named in Article 50(1). However, deployers bear indirect responsibility in several ways:

The EU AI Act has extraterritorial reach under Article 2. If your AI system interacts with people located in the EU, you must comply regardless of where your company is headquartered. A US SaaS company whose chatbot serves European customers is subject to Article 50(1).

3. Systems in Scope: Chatbots, Voice AI, Agents, and More

Article 50(1) applies to any AI system “intended to interact directly with natural persons.” This covers a broad range of systems:

4. The “Obvious” Exception — When Disclosure Is Not Required

Article 50(1) includes an exception: disclosure is not required when the AI nature is “obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and context of use.”

The European Commission’s draft guidelines on Article 50, published in May 2026, recommend interpreting this exception restrictively. The test is objective: would a reasonable person in the actual context of use recognize they are interacting with AI?

When the Exception Likely Applies

A system marketed and branded as an AI assistant (such as “Ask our AI” with a robot icon) would likely qualify. Well-known consumer AI products with established brand recognition — where users deliberately launch an AI interface — would also qualify. The key factor is whether the context provides sufficient signals that AI is involved.

When the Exception Does Not Apply

A chatbot that uses a human name and photo, even if technically AI-powered, does not qualify. A voice system that sounds human without identifying itself as AI does not qualify. An AI that responds in a messaging thread alongside human agents, without distinguishing itself, does not qualify. If there is any ambiguity about whether the user would recognize the AI nature, the safe approach is to disclose.

5. How to Disclose: Design Standards and UX Patterns

Article 50(5) requires that the information be provided “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure,” conforming to applicable accessibility requirements.

The Three Requirements

Clear: The disclosure must be in plain language that a non-technical user can understand. “You are chatting with an AI assistant” is clear. “This interface utilizes large language model technology” is not clear to most users.

Distinguishable: The disclosure must stand apart from surrounding content. It cannot be embedded in a wall of text, hidden in a footer, or presented in the same visual style as regular chat messages. It must be visually or audibly distinct.

At first interaction: The disclosure must appear before or at the moment the user first engages with the AI. Not after the first exchange. Not on a separate page. At the point of interaction.

Recommended UX Patterns for Text-Based Chatbots

Pattern A: Introductory Message. The chatbot’s first message includes a clear statement: “I’m an AI assistant. I can help you with questions about [topic]. How can I help?” This is the simplest and most widely recommended approach.

Pattern B: Persistent Label. A visible label or badge appears in the chat interface at all times — for example, “AI Assistant” next to the chatbot’s name, or a small “AI” badge on every message. This approach is especially useful for chat interfaces where the user may join mid-conversation.

Pattern C: Pre-Interaction Notice. Before the chat window opens, a brief notice appears: “You will be connected to our AI assistant.” The user must acknowledge or proceed. This is more intrusive but provides strong evidence of compliance.

The European Commission and the Code of Practice favor a combination: an introductory message plus a persistent visual indicator throughout the session.

6. Pass/Fail Examples: What Compliance Looks Like

7. Implementation Checklist

8. Special Considerations: Voice AI and Phone Systems

Voice-based AI systems present unique challenges for Article 50(1) compliance because there is no visual interface to display a label. The disclosure must be audible, and it must be delivered in a way that callers can understand and remember.

Requirements for Voice AI Disclosure

Timing: The disclosure must be the first thing the caller hears, or among the first statements. It cannot be delivered after the AI has already begun substantive interaction. “Before I answer that — just so you know, I’m an AI assistant” is too late if the interaction has already progressed.

Clarity: The statement must be spoken at a natural pace, in the language the caller expects, and loud enough to be heard. A rapid, mumbled disclosure fails the “clear” standard.

Persistence: Consider providing periodic reminders during long calls, especially if the voice is very human-like. A caller who is transferred between systems should receive a new disclosure if they are now speaking with AI.

Recommended Script

“Thank you for calling [Company]. You are now speaking with an AI assistant. I can help with [list of capabilities]. If you would like to speak with a person, say ‘human agent’ or press zero at any time.”

9. Special Considerations: AI Email Responders and Agents

AI-powered email systems — from auto-responders that draft replies to fully autonomous email agents — fall under Article 50(1) when they interact with external recipients who may reasonably believe they are communicating with a human.

When Disclosure Is Required

If an AI system generates email responses sent to customers, partners, or the public, and the recipient could reasonably believe a human wrote the reply, disclosure is required. This is especially important when the email is sent from a named account (e.g., “From: Sarah Johnson, Customer Support”) but the content was AI-generated.

Recommended Approach

Include a clear indicator in the email body or signature: “This response was generated by our AI assistant and reviewed by [human/team name].” If the response was fully autonomous (no human review), state: “This response was generated by our AI assistant.”

An alternative is to use a distinct sender identity: “From: [Company] AI Assistant <ai-support@company.com>” rather than a human name.

10. Relationship with GDPR and Other Frameworks

Article 50(1) operates alongside, not instead of, existing transparency requirements. Understanding how these frameworks interact prevents both gaps and redundant compliance efforts.

Article 50(6) explicitly states that the transparency obligations under Article 50 are “without prejudice” to other EU or national transparency requirements. Complying with Article 50(1) does not exempt you from GDPR, and GDPR compliance does not exempt you from Article 50(1). Both must be satisfied independently.

11. Penalties and Enforcement

Under Article 99 of the EU AI Act, violations of Article 50 transparency obligations are subject to administrative fines of up to €15 million or 3% of total worldwide annual turnover in the preceding financial year, whichever is higher. For SMEs and startups, the proportionality principle applies: the lower of the two amounts is the maximum.

Enforcement begins August 2, 2026. National market surveillance authorities designated by each EU Member State will oversee compliance. The European AI Office has direct enforcement authority over general-purpose AI models. Providing incorrect, incomplete, or misleading information to authorities carries a separate fine of up to €7.5 million or 1% of turnover.

While enforcement priorities in the initial period may focus on the most egregious violations, the fining power is operative from day one. Companies that have not implemented disclosure by August 2, 2026 are technically in violation and subject to enforcement action.

12. Frequently Asked Questions

Does this apply if I only use AI internally?

If employees know they are using an AI tool (such as a branded AI assistant or an internal ChatGPT integration), the “obvious” exception likely applies. However, if AI generates responses in internal systems that could be mistaken for human messages (e.g., an AI agent posting in a team chat without identification), disclosure would be required.

What about AI-enhanced search on my website?

A standard search bar that returns AI-generated summaries alongside traditional results should disclose the AI-generated nature of those summaries. The search bar itself is not an “AI system interacting directly with natural persons” — but if it presents AI-generated content as if it were written by humans or curated manually, transparency is needed.

Can I satisfy the requirement with a single disclosure for my entire website?

No. The disclosure must be at the point of interaction — where and when the user actually interacts with the AI system. A site-wide banner saying “We use AI on this website” does not meet the requirement for a specific chatbot unless it is prominently displayed in the chat interface itself.

What if the user already knows it is AI from previous visits?

Article 50(1) requires disclosure “at the time of the first interaction.” Each new session or conversation should include disclosure, even for returning users. A persistent visual label satisfies this automatically for users who return to the same interface.

Does Article 50(1) apply to AI systems that only listen (no response)?

Article 50(1) covers systems “intended to interact directly.” A system that only listens without responding does not interact. However, such systems may fall under Article 50(3) if they perform emotion recognition or biometric categorisation, which has its own disclosure requirement.

What language should the disclosure be in?

The disclosure should be in the language of the interaction. If the chatbot communicates in French, the disclosure should be in French. If the system supports multiple languages, the disclosure should match the user’s selected or detected language. The AI Act does not prescribe specific wording, only that it be “clear and distinguishable.”

Is there a law enforcement exception?

Yes, but narrow. AI systems authorized by law to detect, prevent, or investigate criminal offences are exempt — unless they are publicly accessible for crime reporting. An AI-powered public crime reporting portal must disclose its AI nature. Covert investigative tools do not.

What about AI systems that are partially human, partially AI?

Hybrid systems — where human agents and AI work together — require disclosure when the AI is actively generating responses. If a human agent uses AI to draft a suggested reply but reviews and sends it personally, the interaction is human-led. If the AI generates and sends the response autonomously, it is an AI interaction requiring disclosure.