# Article 50 — EU AI Act Transparency Compliance Kit

**Regulation (EU) 2024/1689 — Article 50: Transparency obligations for deployers of certain AI systems**

*Prepared by HI AI Design, Paris | Last updated: February 2026*

---

## What Article 50 Requires

Article 50 mandates that **any person interacting with an AI system must be informed they are doing so**, unless this is obvious from the circumstances. This applies to:

- Chatbots and conversational AI assistants
- AI-generated content (text, images, audio, video)
- Emotion recognition systems
- Biometric categorisation systems
- Deep fakes / synthetic media

> **Key text:** *"Deployers of an AI system that interacts directly with natural persons shall inform natural persons that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use."*
> — Article 50(1), EU AI Act

---

## Compliance Checklist

### 1. Opening Disclosure (Mandatory)

Every AI chatbot or assistant **must** include a clear disclosure in its first message to the user.

**English template:**
```
I am an AI-powered assistant using a large language model (LLM) to generate my responses. I am not a human.
```

**French template:**
```
Je suis un assistant alimenté par intelligence artificielle utilisant un grand modèle de langage (LLM) pour générer mes réponses. Je ne suis pas un être humain.
```

**Implementation notes:**
- Must appear in the **first message** of every new conversation
- Must appear for **returning visitors** (not just first-time)
- Should be prominent, not buried in fine print
- Should use clear, plain language — avoid jargon

---

### 2. Persistent UI Label (Recommended)

A visible badge or label in the chatbot header that is always visible during the interaction.

**HTML example:**
```html
<span class="ai-badge" title="AI System — EU AI Act Article 50">
    AI Assistant
</span>
```

**CSS example:**
```css
.ai-badge {
    display: inline-flex;
    align-items: center;
    gap: 4px;
    font-size: 10px;
    font-weight: 600;
    letter-spacing: 0.5px;
    text-transform: uppercase;
    color: #f59e0b;
    background: rgba(245, 158, 11, 0.12);
    border: 1px solid rgba(245, 158, 11, 0.3);
    border-radius: 6px;
    padding: 2px 7px;
    width: fit-content;
    cursor: help;
}
```

**Why amber/warning colour?** It's noticeable without being alarming, signals "information" rather than "danger", and remains visible on both light and dark backgrounds.

---

### 3. System Prompt Instructions (Mandatory for LLM-based systems)

Add these rules to the system prompt of any LLM-powered chatbot:

```
EU AI ACT ARTICLE 50 — TRANSPARENCY OBLIGATION (MANDATORY)
- You MUST identify yourself as an AI system at the start of every new conversation.
- If a user asks whether you are human, always respond honestly: you are an AI assistant.
- Never imply, suggest, or allow the user to believe they are speaking with a human.
- Include a brief disclosure in your first message: "I am an AI assistant powered by a large language model."
- This transparency obligation is required by Article 50 of the EU AI Act (Regulation 2024/1689).
```

---

### 4. Returning Visitor Disclosure

Even if the user has chatted before, each new session must include a reminder:

**English:**
```
Reminder: I am an AI assistant, not a human.
```

**French:**
```
Rappel : je suis un assistant IA, pas un être humain.
```

---

### 5. Fallback / Offline Messages

Any fallback or error responses should also identify the system as AI:

```
I'm Sympathico — an AI assistant that helps companies navigate EU AI Act compliance.
Please note: I am an AI system, not a human.
```

---

## Implementation Reference (Sympathico Example)

### Files Modified

| Component | File | What changed |
|-----------|------|-------------|
| Opening greeting (EN/FR) | `SympathicoWidget.js` → `getTranslations()` | Added transparency notice with LLM disclosure |
| Header badge | `SympathicoWidget.js` → `getTemplate()` | Added `.symp-ai-badge` element with "AI Assistant" label |
| Badge styling | `sympathico.css` | New `.symp-ai-badge` class with amber pill design, light/dark theme support |
| Welcome-back greeting | `SympathicoWidget.js` → `getWelcomeBackGreeting()` | Added AI disclosure prefix for returning visitors |
| Fallback messages | `SympathicoWidget.js` → `fallbackGreeting`, `fallbackDefault` | Updated to include AI self-identification |
| System prompt | `sympathico_system_prompt.txt` | Added Article 50 transparency obligation section |

### Greeting Template (Full)

**English:**
```
👋 Hi, I'm [Name] — an AI-powered assistant built by [Company], [City].

🤖 Transparency notice: I am an artificial intelligence system using a large
language model (LLM) to generate my responses. I am not a human. My purpose
is to help you [primary purpose], answer questions about [topic], and connect
you with our human team when needed.

How can I help you?
```

**French:**
```
👋 Bonjour, je suis [Nom] — un assistant alimenté par intelligence artificielle,
créé par [Entreprise], [Ville].

🤖 Avis de transparence : Je suis un système d'intelligence artificielle utilisant
un grand modèle de langage (LLM) pour générer mes réponses. Je ne suis pas un
être humain. Mon rôle est de vous aider à [objectif principal], répondre à vos
questions sur [sujet] et vous mettre en contact avec notre équipe humaine si
nécessaire.

Comment puis-je vous aider ?
```

---

## What Article 50 Does NOT Require

- Disclosing the specific model name (e.g., GPT-4, Claude)
- Disclosing training data sources
- Providing a full technical spec in the chat
- Asking for consent before interacting (that's GDPR, not Article 50)

---

## Penalties for Non-Compliance

| Violation | Maximum Fine |
|-----------|-------------|
| Transparency obligations (Article 50) | **€15M or 3% of global annual turnover** |
| High-risk system obligations | €35M or 7% of global turnover |
| Prohibited practices | €35M or 7% of global turnover |

---

## Timeline

| Date | Milestone |
|------|-----------|
| **1 Aug 2024** | EU AI Act entered into force |
| **2 Feb 2025** | Banned AI practices enforced |
| **2 Aug 2025** | GPAI rules + governance structure apply |
| **2 Aug 2026** | **Full enforcement** including Article 50 transparency obligations |

---

## Quick Integration Guide for Client Websites

### Step 1 — Audit existing chatbot
- Does the first message identify the system as AI? ❑
- Is there a visible "AI" label in the header? ❑
- Does the system prompt instruct the LLM to self-identify? ❑
- Do returning-visitor greetings include AI reminder? ❑
- Do fallback/error messages identify as AI? ❑

### Step 2 — Update greeting message
Copy the template above and customise `[Name]`, `[Company]`, `[City]`, `[primary purpose]`, and `[topic]`.

### Step 3 — Add header badge
Add the HTML badge element to the chatbot header. Apply the CSS styling (adjust colours to match client brand).

### Step 4 — Update system prompt
Add the mandatory transparency obligation block to the LLM system prompt.

### Step 5 — Test all entry points
- New visitor → first greeting ✓
- Returning visitor → welcome-back greeting ✓
- API failure → fallback message ✓
- Direct question "are you human?" → honest answer ✓

### Step 6 — Document
Record the changes in the client's AI governance documentation for audit trail purposes.

---

## Legal References

- **Regulation (EU) 2024/1689** — Artificial Intelligence Act
  - Article 50: Transparency obligations for deployers
  - Recital 132: Transparency for AI interaction
- **ISO/IEC 42001:2023** — AI Management System (supports governance documentation)

---

*This compliance kit is maintained by HI AI Design. For implementation support, contact contact@hiai-design.com*