EU AI Act transparency requirements from 2 August: what businesses need to know

From 2 August 2026, the transparency requirements in Article 50 of the European Union Artificial Intelligence Act begin to apply. They affect businesses that develop AI solutions, introduce AI assistants under their own name, or use certain AI tools in a professional context.

In practice, this does not mean that every email or social media post created with AI assistance must now carry an AI label. The requirements are more specific. They cover direct interaction between people and AI, technical marking of synthetic content, emotion recognition, biometric categorisation, deepfakes, and certain texts on matters of public interest.

This article explains how to understand your company’s role and what to check before and after 2 August.

The short answer. A company should first identify where and how it uses AI. It must then determine whether it is the provider or deployer of the AI system in each case. The next step is to introduce the right disclosures, content review, and demonstrable accountability.

The AI Act is already in force, but new requirements now begin to apply

The EU AI Act entered into force on 1 August 2024 and is being applied in stages. From 2 August 2026, the transparency rules in Article 50 begin to apply.

This distinction matters. In the summer of 2026, some requirements for high-risk AI systems were moved to later dates. The new deadlines for certain high-risk systems are 2 December 2027 and 2 August 2028. This is not a general postponement of the AI Act, and it does not remove the Article 50 transparency obligations that apply from 2 August 2026.

In July, the European Commission published detailed guidelines and practical examples. They help businesses determine which organisations and use cases fall within the scope of the requirements.

First determine whether your company is a provider or a deployer

A company’s obligations under the AI Act depend on its role in a particular solution.

Provider of an AI system

A provider develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. For example, a business may become a provider if it creates a branded customer service AI agent and makes that solution available to customers.

The provider must ensure that an interactive AI system is designed so people are clearly informed that they are interacting with AI. In certain cases, the provider of a generative AI system must also ensure that generated content carries a machine-readable mark.

Deployer of an AI system

A deployer is a company or other organisation that uses an AI system under its authority in a professional context. Individual employees, contractors, and content creators are generally not separate deployers when they act on the company’s instructions and under its control. The responsible party remains the company.

A business can have different roles across different projects. It may be a deployer when using an off-the-shelf AI tool. If it offers a significantly customised AI system under its own name, it may become a provider. The role should therefore be assessed separately for every solution.

Four situations in which transparency is especially important

1. A person interacts directly with AI

When an AI system conducts a genuine two-way exchange with a person, that person must be informed that they are interacting with AI. This can apply to a website chatbot, voice agent, virtual adviser, or customer support assistant.

The information must be clear from the beginning of the first interaction. A simple disclosure could say “You are speaking with an AI assistant”. An additional notice may not be necessary when the interaction with AI is already obvious to an average person, but the European Commission recommends interpreting this exception narrowly.

A well-designed solution should also provide a way to hand the conversation to a person when necessary, especially when the agent handles payments, complaints, contracts, or other important matters.

2. AI generates or substantially modifies content

Providers of AI systems must, in certain cases, mark synthetic text, images, audio, or video in a machine-readable format. This allows technical systems to detect that content was generated or manipulated using AI.

This requirement primarily applies to the provider of the system, not to every employee using an off-the-shelf tool. A business acting as a deployer should nevertheless check whether its chosen supplier provides the required technical marking and whether that marking is accidentally removed during content processing or publication.

There are exceptions. Machine-readable marking is not intended for source code, purely technical machine-to-machine communication, or standard editing functions that do not substantially change the meaning of the original content. The guidelines also envisage a narrow exception for certain closed business-to-business or industrial uses.

3. Emotion recognition or biometric categorisation is used

If a company uses AI to recognise emotions or perform biometric categorisation, people exposed to the system must be informed that it is operating. This applies to both real-time use and later analysis of recordings.

Such a solution requires particularly careful legal and ethical assessment. A notice alone may not be enough, as other AI Act restrictions, data protection requirements, and employment rules may also apply.

4. A deepfake or AI text on a matter of public interest is published

An AI-generated or manipulated image, audio recording, or video may be a deepfake when it convincingly resembles a real or plausibly existing person, object, place, or event and can mislead people about its authenticity. Such content must be clearly disclosed no later than the first exposure. An invisible technical mark alone is not sufficient.

A separate requirement applies to AI-generated or substantially manipulated text published to inform the public about matters of public interest. Examples can include politics, public services, public health, security, the environment, and significant economic or financial developments.

If a competent person substantively reviews the text before publication and the organisation assumes editorial responsibility, an AI label is not required under Article 50. A grammar check or purely formal approval is not considered sufficient editorial control.

Does every business post created with AI need a label

No. Article 50 of the AI Act does not create a universal obligation to label all content produced with AI assistance.

• An internal email draft reviewed and sent by an employee will generally not fall under the labelling requirement for public-interest text.
• A product description or social media post is not automatically subject to labelling simply because AI helped write it.
• For text on a matter of public interest, substantive professional review and clear editorial responsibility are important.
• A realistic image is not automatically a deepfake. The key question is whether it resembles an existing or plausibly existing subject and can mislead people about authenticity.
• An AI system that operates only in the background and does not communicate directly with a person is outside the requirement to disclose direct interaction with AI.

A company should not rely only on the name or marketing description of a tool. It needs to assess what the system actually does, who sees the output, and who assumes responsibility for the publication or decision.

An important exception until 2 December 2026

A limited transition period until 2 December 2026 applies only to the machine-readable marking and detection requirement for generative AI systems placed on the market before 2 August 2026.

This is not a general four-month extension for all transparency obligations. The requirement to inform people about direct interaction with AI and the relevant deployer obligations apply from 2 August.

Content generated before 2 August does not have to be labelled retroactively. The European Commission nevertheless encourages voluntary labelling where it is practical.

A practical action plan for businesses

1. Create an inventory of AI uses. Include customer chatbots, voice agents, content tools, document analysis, recruitment tools, and AI features built into other systems.
2. Determine the role for every solution. Record whether the company is the provider, the deployer, or both.
3. Check the user interface. If a person communicates directly with AI, add a clear notice at the beginning of the interaction.
4. Define human involvement. Decide which results need review, which actions require approval, and when a conversation or task must be handed to a person.
5. Review the content publication process. Identify who performs substantive fact checking and who assumes editorial responsibility.
6. Review suppliers and integrations. Confirm whether the AI tool provides technical marking, audit records, and the necessary documentation.
7. Keep evidence. Document the tools used, disclosure wording, responsible people, approval steps, and implemented changes.

This inventory does not need to begin as a complex legal project. A shared table or an internal system page can be enough at first, as long as every AI use has an owner, purpose, data sources, risks, and controls.

What this means when building AI systems

Transparency is easier and less expensive to design into a system than to add after launch. A customer chatbot needs more than a disclosure line. It may also need user role checks, access restrictions, approval before significant actions, handover to a person, and an audit trail.

A content system can also include a clear workflow. AI prepares a draft, a responsible specialist checks the facts and sources, an editor approves publication, and the system stores the review history. This supports compliance while also improving content quality.

VIZUAL builds AI integrations and business systems that can include disclosures, human approval points, access permissions, audit trails, and connections to a company’s CRM, ERP, and other systems.

If your company already uses an AI assistant or is planning a new AI integration, review transparency before scaling the solution. Contact VIZUAL to discuss the system design and practical control points.

---

Official sources

• European Commission guidelines on the transparency obligations in Article 50 of the AI Act
• European Commission questions and answers on Article 50
• Quick facts on transparency rules for AI systems
• Council of the EU timeline for the AI Act

This article was prepared on 31 July 2026 and provides general information. It is not legal advice. The compliance of a specific AI solution should be assessed with appropriate legal and data protection specialists.