---
title: What Businesses Must Do After 2 August 2026 - EU AI Act Article 50
date: 2026-08-16T13:22:16Z
modified: 2026-08-27T13:23:12Z
permalink: "https://meganova.agency/ai-enterprise-systems/what-businesses-must-do-after-2-august-2026-eu-ai-act-article-50/"
type: post
status: publish
excerpt: ""
wpid: 21268
categories:
  - AI And Enterprise Systems
tags:
  - AI And Enterprise Systems
  - AI Search Optimization
  - AI Transparency
featured_image: "https://meganova.agency/wp-content/uploads/2026/08/eu_compliance_hologram.jpg"
featured_image_alt: EU AI Act Article 50
timestamp: 2026-08-27T13:23:12Z
---

- **The preparation phase is over.** Article 50 of the[ EU AI Act](https://meganova.agency/wp-content/uploads/wp-mfa-exports/page/ai-search-visibility.md) took effect on **2 August 2026**. Enforcement has begun. The question is no longer “when will we need to comply?” but “are we compliant now?” [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)
- **Four distinct transparency obligations now apply.** Providers and deployers of AI systems must: (1) disclose AI interaction, (2) mark synthetic content as machine-readable, (3) inform individuals about emotion recognition/biometric categorisation, and (4) label deepfakes and certain AI-generated public-interest text. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)[](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)
- **A narrow transition exists — but only for Article 50(2).** Systems already placed on the market before 2 August 2026 have until **2 December 2026** to implement machine-readable marking. Everything else applies in full, effective immediately. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)[](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

---

## What Changed on 2 August 2026

On 2 August 2026, the transparency obligations of[ **Article 50 of the EU AI Act**](https://meganova.agency/wp-content/uploads/wp-mfa-exports/post/eu-ai-act-article-50-explained.md) became law in force, not a date to work towards. This is a fundamental shift: “Until last week an organisation could say it was preparing. From now on the question is whether the disclosure is there.” [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

The enforcement structure the AI Act prescribes came into operation on the same day. National market surveillance authorities can now use their enforcement powers. For the transparency duties, a breach can lead to fines of up to **€15 million or 3% of total worldwide annual turnover**, whichever is higher. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

### Article 50’s Four [Transparency](https://meganova.agency/wp-content/uploads/wp-mfa-exports/post/how-ai-search-is-changing-brand-discovery-in-2026.md) Obligations

Article 50 contains four [transparency duties](https://meganova.agency/wp-content/uploads/wp-mfa-exports/post/best-ai-disclosure-plugins-for-wordpress-an-honest-comparison.md), split between providers and deployers. All four have applied since 2 August 2026. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)[](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)



| Provision | What | Who Must Comply |
| --- | --- | --- |
| **50(1)** | Make clear that a person is interacting with an AI system | Provider |
| **50(2)** | Mark generated audio, image, video or text in a machine-readable format | Provider |
| **50(3)** | Inform people when emotion recognition or biometric categorisation is applied | Deployer |
| **50(4)** | Disclose deepfakes and certain AI text of public interest | Deployer |

> _**Critical distinction:** Article 50 is not a high-risk regime. There is no conformity assessment, no Annex IV technical documentation, and no registration in an EU database. That does not make it optional, but it does make the task bounded: four concrete behaviours, not a full management system._ [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

## The Four Obligations in Detail

### Article 50(1): Direct AI Interaction

**Who must comply:** Providers of AI systems intended to interact directly with natural persons.

**The requirement:** Users must be informed, no later than the first interaction, that they are interacting with an AI system. This applies to chatbots, AI assistants, AI agents, and other conversational interfaces. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Format:** Textual, auditory, visual, or multi-modal formats — provided the person is properly and effectively informed without risk of deception or manipulation. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Exceptions:**

- Where the AI nature of the interaction is “obvious” to a reasonably well-informed, observant member of the target audience. This is construed **narrowly**; providers must be able to demonstrate the obviousness. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)
- Where the AI system is authorised to detect, prevent, investigate, or prosecute criminal offences. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Practical gap:** A field test of ten Dutch chatbots found that only three explicitly told users they were talking to AI. Two called themselves “chatbot” without using the word “AI”, and three chose softer labels such as “digital assistant.” Until 2 August, those were design choices. Now they are non-compliance. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

### Article 50(2): Marking and Detection of Synthetic Content

**Who must comply:** Providers of AI systems that generate or manipulate synthetic audio, image, video, or text content.

**The requirement:** This is a two-fold obligation: [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

1. **Mark in a machine-readable format** — outputs must carry machine-readable marks at the time of generation, enabling automated detection of synthetic origin. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)
2. **Provide a means for detection** — natural persons must be able to identify and distinguish AI-affected content. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Transitional period:** Systems already placed on the market before 2 August 2026 have until **2 December 2026** to implement the marking and detection obligation. Systems placed on the market after 2 August 2026 must have the marking in place immediately. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)[](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

**What is covered:** Synthetic audio, image, video, or text outputs, including virtual/augmented reality content and content mixed with human-created material. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**What is exempt:** Standard editing (spell-checking, grammar correction, basic translation, formatting changes, noise removal). [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

### Article 50(3): Emotion Recognition and Biometric Categorisation

**Who must comply:** Deployers of AI systems used for emotion recognition or biometric categorisation.

**The requirement:** Individuals must be informed when they are exposed to AI systems used for these purposes. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Context:** Emotion recognition uses AI to infer a person’s emotions based on biometric data. Biometric categorisation involves aligning people to specific categories based on biometric data. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

**Additional safeguards:** Emotion recognition systems are classified as high-risk under the AI Act. This transparency obligation operates alongside other high-risk AI safeguards. Any processing of biometric data falls within the scope of the GDPR. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

### Article 50(4): [Deepfakes and Public-Interest Text](https://meganova.agency/wp-content/uploads/wp-mfa-exports/post/how-to-label-ai-generated-content-on-your-website-text-images-video.md)

**Who must comply:** Deployers who use AI systems to produce or modify video, image, or audio content constituting a deepfake, or who publish AI-generated text on matters of public interest.

**The requirement:** Deployers must clearly disclose: [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)[](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)

- **Deepfakes:** Where content is artificially generated or manipulated in a way that resembles a real person, place, or event.
- **AI-generated public-interest text:** Where text generated by AI is published with the purpose of informing the public on matters of public interest (including news and election-related material). [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**The Code of Practice provides an optional EU icon** in three variants that deployers may use to satisfy the labelling obligation in a standardised way. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**Human review exception:** The public-interest text obligation does **not** apply where the AI-generated content has undergone a process of **human review or editorial control** and a natural or legal person holds editorial responsibility for the publication. Such checks must be substantive, not limited to superficial matters or cursory approval.

**Carveouts:** Artistic, creative, satirical, or fictional works where human editorial review and responsibility are exercised. These are narrow—they apply only where the creative or fictional nature of the content is apparent from context. Deepfake elements in non-satirical contexts are unlikely to qualify. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

## The 2 August → 2 December Transition Window

The transitional period is narrower than many assume. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)



| Provision | Applies From | Transition? |
| --- | --- | --- |
| 50(1) — AI interaction disclosure | 2 August 2026 | No |
| 50(2) — Machine-readable marking | 2 August 2026 | **Yes, until 2 December 2026** (systems placed on market before 2 August) |
| 50(3) — Emotion recognition disclosure | 2 August 2026 | No |
| 50(4) — Deepfake/public-interest text disclosure | 2 August 2026 | No |

> _**Key takeaway:** Only Article 50(2), the machine-readable marking of synthetic output, has time until 2 December 2026, and that extension applies solely to systems already placed on the market before 2 August 2026. The visible disclosure of deepfakes under paragraph 4 does **not** fall under the transitional period at all._ [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

_**The next hard date** is 2 December 2026._ [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

## The Code of Practice on Transparency of AI-Generated Content

The European Commission published the **Code of Practice on Transparency of AI-Generated Content** in June 2026. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**Key features:**

- The Code implements the mandatory transparency obligations under Articles 50(2) and (4) of the AI Act. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)
- It is **voluntary in adoption** but translates legally binding AI Act obligations into practical implementation guidance. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)
- A firm may meet Article 50 requirements by following the Code **or by a demonstrably equivalent alternative approach.** [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**The Code distinguishes:**

- **Fully AI-generated content:** Content fully and autonomously generated by the AI system. [](https://digital-strategy.ec.europa.eu/ro/news/working-groups-advance-discussions-transparency-obligations-under-article-50-ai-act)
- **AI-assisted content:** Content involving a mix of human and AI involvement (including relatively minor AI alterations). [](https://digital-strategy.ec.europa.eu/ro/news/working-groups-advance-discussions-transparency-obligations-under-article-50-ai-act)

**Signatories** can rely on the Code to demonstrate compliance with Articles 50(2), (3), and (5). Signing remains possible, even though the initial signatories list has closed. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

## If You Operate a Website with AI Chatbots, AI-Generated Content, or AI Assistants

Article 50(1) applies directly. Your chatbot, AI assistant, or conversational interface must clearly disclose that users are interacting with AI. This is not optional. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

### If You Use AI to Generate Content (Text, Images, Video, Audio)

Article 50(2) and (4) apply. You must mark AI-generated content in a machine-readable format and disclose deepfakes and public-interest text. The Code of Practice provides practical implementation guidance. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

### If Your Organisation Provides AI Systems to Others

You must ensure your systems include the required disclosures and marking mechanisms. Many organisations are the provider for one system and the deployer for another—the analysis must run per system. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

### The Enforcement Reality

National market surveillance authorities can now use their powers. A breach can lead to fines of up to €15 million or 3% of total worldwide annual turnover. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

> _“The practical consequence is that demonstrability now counts. A supervisor asking how you meet Article 50 is not asking about your intentions but about your systems: which ones communicate with people, which generate synthetic content, where the disclosure sits, and who established that and when.”_ [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

## Frequently Asked Questions (FAQ)

**Q: Does Article 50 apply to my website if I’m not in the EU?**
A: Yes, if you deploy AI systems or place AI systems on the EU market that affect EU citizens or businesses.

**Q: Does Article 50 apply to AI-assisted content?**
A: The distinction between AI-generated and AI-assisted content matters. The Code of Practice provides a taxonomy. If the content is “AI-generated” (fully or substantially autonomous), disclosure is required. “AI-assisted” (human with AI involvement) may have different considerations. [](https://digital-strategy.ec.europa.eu/ro/news/working-groups-advance-discussions-transparency-obligations-under-article-50-ai-act)

**Q: What happens if we don’t comply?**
A: Fines of up to €15 million or 3% of total worldwide annual turnover. Enforcement started on 2 August 2026. [](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

**Q: Do we need to comply if our AI system was placed on the market before 2 August?**
A: Yes. The only transition is for Article 50(2) machine-readable marking, which has until 2 December 2026. All other provisions apply immediately. [](https://publicationportalpreview.hlc.com/en/publications/the-european-commission-issues-draft-guidelines-on-the-transparency-requirements-under-the-ai-act)[](https://www.aiactblog.nl/en/posts/article-50-applies-now-what-changed)

**Q: What is the Code of Practice and do we need to follow it?**
A: It’s a voluntary framework that provides practical guidance. You can follow it or use a demonstrably equivalent alternative approach. Following the Code provides a safe harbour for compliance demonstration. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)

**Q: What is the EU icon?**
A: The Code of Practice provides an optional EU icon in three variants that deployers may use to satisfy the labelling obligation in a standardised way. [](https://www.licentium.io/post/eu-code-of-practice-ai-generated-content-transparency-2026)