---
title: "EU AI Act Article 50: What Support Chatbots Must Disclose"
description: "EU AI Act Article 50 has applied since 2 August 2026. See what your support chatbot must disclose, who’s responsible, and how to comply."
url: "https://www.bolddesk.com/blogs/eu-ai-act-article-50"
last_updated: 2026-11-09
---

-   Blogs [Explore Blogs Blogs](https://www.bolddesk.com/blogs)

# EU AI Act Article 50: What Your Support Chatbot Must Disclose

-    [by Susan Wairimu Mwangi](https://www.bolddesk.com/author/susan-wairimu-mwangi)
-   Updated on: October 23, 2026

![EU AI Act Article 50 illustration showing AI assistant, chatbot, user interaction, compliance document, and security shield](https://www.bolddesk.com/wp-content/uploads/2026/10/EU-AI-Act-Article-50-banner-1024x546.webp)

Level Up Your Customer Support with BoldDesk

[![Sign up for BoldDesk](https://www.bolddesk.com/wp-content/uploads/2026/09/80ada21c-bolddesk-blog-ads.svg)](https://app.bolddesk.com/signup?)

[Free Trial](https://app.bolddesk.com/signup?)

[Book Demo](https://www.bolddesk.com/request-demo)

Summarize with:

Share

> **TL;DR:** Since 2 August 2026, Article 50 requires disclosure of AI interactions. The Digital Omnibus did not delay this requirement. Generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to meet Article 50(2). Fines can reach €15 million or 3% of worldwide annual turnover.

If your customer support chatbot uses AI, customers should not have to guess whether they are speaking to a person or a machine.

[EU AI Act Article 50](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50) sets transparency rules for AI systems that talk directly to people, including chatbots and [AI agents in customer service](https://www.bolddesk.com/blogs/ai-agents-in-customer-service).

For support teams, that makes AI disclosure more than a UX decision. Since 2 August 2026, teams must understand when disclosure applies and what providers and deployers must do.

This guide explains how EU AI Act Article 50 applies to customer support, including AI disclosure requirements and the different responsibilities of providers and deployers.

## What does EU AI Act Article 50 require?

EU AI Act Article 50 sets transparency requirements for specific AI systems and AI-generated content. For customer support teams, Article 50(1) requires providers to ensure users know they are interacting with AI, unless this is already obvious.

However, Article 50 of the AI Act covers more than chatbots. Requirements vary by the type of AI system, the content involved, and whether the obligation falls on the provider or deployer.

![EU AI Act Article 50 transparency obligations: AI disclosure, content marking, emotion recognition, and deepfake labeling](https://www.bolddesk.com/wp-content/uploads/2026/10/EU-AI-Act-Article-50-transparency-requirements.webp)

The table below summarizes the four transparency obligations under Article 50.

| **Provision** | **Who is responsible?** | **What does it cover?** |
| --- | --- | --- |
| **AI interaction disclosure (Article 50(1))** | Provider | AI systems designed to interact directly with people must inform users they are interacting with AI, unless this is obvious to the intended audience. |
| **Marking AI-generated synthetic content (Article 50(2))** | Provider | AI systems, including general-purpose AI, that generate synthetic content must mark outputs so they are detectable as artificially generated or manipulated. |
| **Emotion recognition and biometric categorization (Article 50(3))** | Deployer | People exposed to emotion-recognition or biometric-categorization systems must be informed that those systems are being used. |
| **Deepfake and public-interest text labeling (Article 50(4))** | Deployer | Certain deepfakes and AI-generated or manipulated text published to inform the public on matters of public interest must be disclosed, subject to specific exceptions. |

These distinctions come directly from the EU AI Act and European Commission’s Guidelines on [transparency obligations for providers and deployers](https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems) of AI systems.

For public-interest text under Article 50(4), disclosure is unnecessary if humans review the content and accept editorial responsibility.

## Does EU AI Act Article 50 apply to customer support chatbots?

Article 50(1) generally applies when an AI system is designed to talk directly with customers in a two-way exchange. This includes conversational systems such as chatbots and [AI support agents](https://www.bolddesk.com/support-agent).

It does not cover AI that runs only in the background, talks only to other machines, or passes content through a human.

For example, say AI drafts a reply and an agent reviews and sends it. The customer is dealing with the agent, not directly with the AI.

That means the direct-interaction requirement in Article 50(1) is not triggered by that exchange. Other legal obligations may still apply depending on the use case.

If you’re unsure where a chatbot ends and a more autonomous AI agent begins, see our guide to [AI agent vs. chatbot](https://www.bolddesk.com/blogs/ai-agent-vs-chatbot).

## Key EU AI Act dates support teams should know

The Digital Omnibus on AI did not delay Article 50 as a whole. Article 50 has applied since 2 August 2026.

There is one short extension. Generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking under Article 50(2).

| **Date** | **What it means** |
| --- | --- |
| **2 August 2026** | Article 50 transparency requirements apply. |
| **2 December 2026** | Article 50(2) machine-readable marking applies to generative AI systems already on the market before 2 August 2026. |
| **2 December 2027** | The postponed rules for Annex III high-risk AI systems apply under the Digital Omnibus. |

The final Digital Omnibus is [Regulation (EU) 2026/1744](https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng), published on 24 July 2026 and in force since 27 July 2026.

**Note:** 2 December 2026 is not a general Article 50 grace period. It applies only to Article 50(2) requirements for eligible generative AI systems already on the market.

## When and how should a support chatbot disclose that it is AI?

For Article 50(1), users must be told from the start that they are talking to AI. Under Article 50(5), the notice must be clear, distinguishable, accessible, and provided no later than the first interaction or exposure.

Generic notices such as “this website uses AI” are not enough. Technical descriptions such as “this system uses LLMs” also do not clearly tell users they are interacting with AI.

In practice, the disclosure does not need to be lengthy. A short, plain introduction can clearly communicate the AI’s role.

**Examples by support channel:**

-   **Web and in-app chat:** “Hi, I’m an AI support assistant. How can I help you today?”
-   **Named AI persona:** “Hi, I’m Nova, an AI support assistant for \[Company\]. How can I help?”
-   **WhatsApp and messaging:** “Hi, this is \[Company\]’s AI support assistant. How can I help with your request?”
-   **AI-sent email:** “This message was generated and sent by \[Company\]’s AI support assistant.”
-   **Voice support:** “Hello, you’re speaking with \[Company\]’s AI support assistant.”

In long chats, complaints, billing disputes, or conversations with vulnerable customers, repeat the AI notice when appropriate. As a best practice, the system should confirm it is AI whenever a customer asks.

A persona name such as Nova should not replace a clear AI label. Show the notice to the customer directly. Don’t rely only on a robot icon, a general website notice, text on another page, text hidden in a footer, or terms and conditions.

When AI and humans share a chat, make each switch clear. If an AI agent hands the chat to a human, tell the customer a person is taking over. If AI takes back the chat, say so.

Clear identification also helps customers understand how an [AI-to-human handoff](https://www.bolddesk.com/blogs/ai-to-human-handoff) changes who is handling their conversation.

## Provider vs. deployer: Who is responsible for Article 50 compliance?

EU AI Act Article 50 gives different transparency duties to providers and deployers. For support teams, knowing your legal role shows which rules apply.

Deployer duties cover emotion recognition and biometric categorization under Article 50(3), and certain deepfakes and public-interest AI-generated content under Article 50(4).

| **Scenario** | **What it means for a support team** |
| --- | --- |
| **Vendor’s built-in AI agent** | The vendor is generally the provider of the AI system it places on the market under its name. Check that the product provides the required disclosure and do not configure the customer experience in a way that hides it. |
| **Your own or rebranded system** | If your company builds, commissions, or rebrands an AI system, including one built on open-source models, it may qualify as the provider. In that case, it may assume the associated Article 50 responsibilities. |
| **AI-assisted draft responses** | Where AI creates a draft that a human meaningfully reviews and sends, there is generally no direct customer-AI interaction. If drafts are sent automatically, customers may effectively interact directly with AI. |

**Key takeaway:** Don’t assume “we bought the chatbot from a vendor” settles every compliance question. Determine who legally qualifies as the provider, which Article 50 duties they have, and whether your setup creates other duties.

### Check sentiment and emotion features

Some support platforms analyze customer emotion, such as voice-based stress detection during calls.

If a system infers emotions from biometric data such as voice or facial features, Article 50(3) requires disclosure. The General Data Protection Regulation (GDPR) may also apply.

Using emotion recognition on employees is different. Article 5(1)(f) of the EU AI Act generally prohibits emotion-recognition systems in workplace settings, subject to limited exceptions for medical or safety purposes.

Text-only sentiment analysis is usually treated differently, so seek legal advice for specific tools.

## How support teams can prepare for Article 50 compliance

Start by mapping where AI directly interacts with customers, including chat, messaging, voice, autonomous email, and other customer-facing channels.

Then proceed as shown below:

-   **Identify your legal role:** Check whether your company is a provider, deployer, or potentially both for each AI system.
-   **Review AI disclosures:** Make sure users are informed when they are talking to AI where required. Pair persona names with “AI,” provide disclosures in the languages you support, and meet relevant accessibility requirements.
-   **Test every channel:** Check that the notice shows up across chat, messaging, AI-sent email, voice, and AI-to-human handoffs.
-   **Audit additional AI features:** Check if AI-made content, emotion recognition, or biometric categorization adds more Article 50 duties. Basic editing tools, such as grammar correction, are generally exempt from marking.
-   **Check agent-assist workflows:** Ensure automation does not bypass meaningful human review of AI-generated drafts. Automatic sending may trigger Article 50(1) disclosure requirements.
-   **Clarify responsibilities:** Define responsibility for transparency obligations and document it in vendor agreements.
-   **Keep records:** Keep notes on the AI systems you use, what they do, and the controls you have in place.
-   **Retest after changes:** Recheck disclosures whenever you change personas, greetings, channels, AI features, or automation workflows.

## Evaluating customer support platforms for EU AI Act transparency

When you compare AI support platforms, don’t judge only answer quality and automation. Ask how their design helps you comply with EU AI Act Article 50.

Look for:

-   **Clear AI identification:** Can customers easily recognize the interaction is AI-powered?
-   **First-interaction disclosure:** Can the system tell customers from the start that they are talking to AI?
-   **Accessibility:** Can disclosures be presented clearly and accessibly?
-   **Multilingual deployment:** Can the appropriate disclosure be shown wherever the AI serves customers in different languages?
-   **Channel coverage:** Can the AI label stay visible across chat, messaging, voice, and other channels?
-   **Human handoff:** Can the experience make AI-to-human transitions clear?
-   **Synthetic-content controls:** If Article 50(2) applies, how does the provider meet machine-readable marking requirements? Has the vendor signed the EU Code of Practice on Transparency of AI-Generated Content? If not, what evidence supports compliance?
-   **Provider documentation:** Does the vendor explain which obligations it handles and give customers enough information to assess their own use?

If you’re comparing platforms more broadly, our guide to the [best AI customer service software](https://www.bolddesk.com/blogs/ai-customer-service-software) covers AI capabilities, integrations, pricing, scalability, and support use cases.

## Building transparent AI support under the EU AI Act Article 50

EU AI Act Article 50 makes transparency a core requirement for AI-powered customer support. Support teams should identify customer-facing AI and understand their duties before using or changing these systems.

With [BoldDesk](https://www.bolddesk.com/), customers can request human assistance during an AI conversation, while agents can review AI-generated draft responses before they are sent.

Organizations can customize AI agent greetings and customer-facing messaging, helping make AI interactions clearer to customers. Start your [15-day free trial](https://app.bolddesk.com/signup/) to explore BoldDesk’s AI-assisted customer support capabilities.

**Disclaimer:** This article is provided for general informational purposes only. It does not constitute legal advice. Consult qualified legal counsel regarding EU AI Act compliance.

**Related articles**

-   [Human-in-the-Loop AI for Customer Service: When to Use It](https://www.bolddesk.com/blogs/human-in-the-loop-ai-for-customer-service)
-   [How to Prevent AI Hallucinations in Customer Service and Build Trust](https://www.bolddesk.com/blogs/ai-hallucinations-in-customer-service)
-   [8 Practical Ways to Use Generative AI in Customer Service](https://www.bolddesk.com/blogs/generative-ai-in-customer-service)

## FAQs

### Are there any exceptions to Article 50’s AI transparency requirements?

Yes, but the exception is narrow. Disclosure is unnecessary only when AI involvement is obvious to a reasonably well-informed, observant, and circumspect (careful) user. The assessment should reflect an average member of the system’s intended audience.

### Is a customer support chatbot automatically a high-risk AI system?

Not simply because it is an AI chatbot. Article 50 transparency requirements are separate from high-risk classification.

High-risk status depends on how the system is used. Some applications, such as creditworthiness assessments, may fall under Annex III high-risk requirements. Following the Digital Omnibus changes, those obligations apply from 2 December 2027.

### Does Article 50 of the AI Act apply to non-EU companies?

Yes, it can. The [European Commission’s guidance](https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act) explains that the Act can cover providers outside the EU when their AI system’s output is used in the EU.

Companies should consider where their AI output is used, not just where they are based or registered.

### What are the common Article 50 compliance mistakes to avoid?

Common mistakes include unclear or late AI notices, assuming customers will spot AI on their own, ignoring accessibility, and mixing up provider and deployer duties.

### What happens if a business does not comply with Article 50 of the AI Act?

Breaking the rules can lead to enforcement action and fines.

According to Article 99 of the AI Act, Article 50 breaches can trigger fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and startups, the applicable cap is whichever amount is lower.

ABOUT THE AUTHOR

[![Picture of Susan Wairimu Mwangi](https://secure.gravatar.com/avatar/8fbce27abf05e61a96a5639c63402ee82e91330de9dabd8a6bfb29449f4d3826?s=300&d=mm&r=g)](https://www.bolddesk.com/author/susan-wairimu-mwangi)

[Susan Wairimu Mwangi](https://www.bolddesk.com/author/susan-wairimu-mwangi)

Susan is a creative content strategist at Syncfusion, crafting engaging stories that connect brands with audiences. She blends insight, clarity, and creativity to bring ideas to life while elevating content that enhances customer support and overall customer experience.

[Read more articles by this author](https://www.bolddesk.com/author/susan-wairimu-mwangi) 

Be the first to get updates

## Explore more content

[![AI customer service agent using analytics and automation tools to improve support performance.](https://www.bolddesk.com/wp-content/uploads/2026/09/AI-customer-service-business-case-1024x546.webp)](https://www.bolddesk.com/blogs/ai-customer-service-business-case)

### [The AI Customer Service Business Case: What 2026 Research Supports](https://www.bolddesk.com/blogs/ai-customer-service-business-case)

-   [by Gibson Amisi Kashi](https://www.bolddesk.com/author/gibson-amisi-kashi)
-   September 21, 2026

[![AI agents for HR and employee support automating onboarding, approvals, records, and employee self-service](https://www.bolddesk.com/wp-content/uploads/2026/09/ai-agents-for-hr-and-employee-support-banner-image-1024x546.webp)](https://www.bolddesk.com/blogs/ai-agents-for-hr-and-employee-support)

### [AI Agents for HR and Employee Support: Benefits and Use Cases](https://www.bolddesk.com/blogs/ai-agents-for-hr-and-employee-support)

-   [by Sarah Adhiambo Ogada](https://www.bolddesk.com/author/sarah-adhiambo-ogada)
-   September 18, 2026

[![How to reduce ticket reassignments through smart ticket distribution and workflow optimization](https://www.bolddesk.com/wp-content/uploads/2026/09/How-to-reduce-support-tickets-reassignments-1024x546.webp)](https://www.bolddesk.com/blogs/how-to-reduce-ticket-reassignments-webinar)

### [Webinar Recap: Cut Ticket Reassignments with Clear Ownership](https://www.bolddesk.com/blogs/how-to-reduce-ticket-reassignments-webinar)

-   [by Maxwel Odhiambo](https://www.bolddesk.com/author/maxwel-odhiambo)
-   September 17, 2026
