The EU AI Act will introduce new transparency duties for businesses that provide or deploy certain AI systems from 2 August 2026. Article 50 requires disclosures when people interact with AI and creates labeling expectations for AI-generated or manipulated content, including deepfakes. For companies using chatbots, generative media tools, emotion-recognition systems, or biometric categorisation, the practical task is to identify where AI is visible to people and make its role clear.
The obligations are part of the official text of Regulation (EU) 2024/1689, commonly known as the EU AI Act. The rules are intended to reduce deception and help people recognise when content or an interaction involves AI. They also create a more concrete implementation deadline for tool providers and organisations that put AI into customer, marketing, editorial, or public-facing workflows.
What Article 50 requires
Article 50 divides transparency obligations between providers, which develop or place AI systems on the market, and deployers, which use relevant systems. The exact duty depends on the type of system and how it is used.
| Situation | Who has the duty | Required transparency action |
|---|---|---|
| A person interacts with an AI system | Provider | Inform individuals that they are interacting with AI. |
| An AI system generates or manipulates content | Provider | Make the artificial origin of content verifiable through machine-readable marking. |
| Emotion recognition or biometric categorisation is used | Deployer | Inform people exposed to the system. |
| Deepfakes or AI-produced public-interest text are published | Deployer | Disclose that the content was artificially generated or manipulated. |
The distinction matters because a business may hold both roles in different circumstances. A company that buys a customer-support AI tool will usually focus on its deployment and user-facing disclosures. A software business that builds its own chatbot, content generator, or AI media product may also need to address provider duties, including mechanisms for marking output.
The rules are not a blanket requirement to label every use of software containing AI. They focus on defined circumstances where a person is interacting with AI, where generated or manipulated content should be detectable, or where systems can affect how people are perceived through emotion recognition or biometric categorisation. Businesses should assess actual user journeys and outputs rather than rely on broad descriptions from vendors.
Content labels and deepfake disclosures
The most visible changes may concern synthetic content. Providers of systems that generate or manipulate audio, images, video, or text have duties to make the origin of relevant content detectable in a machine-readable form. That is different from merely adding a caption to an image or a line of text below a video. Machine-readable marking is intended to support technical verification of artificial origin.
Deployers have their own disclosure duties when they publish deepfakes or AI-generated public-interest text. The EU has also published a voluntary Code of Practice on Marking and Labelling AI-Generated Content. It offers a practical framework for providers and deployers, encourages open standards for content marking, and includes an EU icon set intended to support consistent labeling.
The code is voluntary, but it can help businesses move from a high-level legal obligation to repeatable product and publishing practices. For example, a provider can map how its tool records content provenance, while a marketing or editorial team can define when visible disclosures are required before publishing AI-assisted material.
Preparing for the August 2026 deadline
The central date is 2 August 2026, when the Article 50 obligations are set to apply. There are transitional considerations for certain generative AI content created by systems placed on the market before that date. The marking obligation for that older content may extend to December 2026. Businesses relying on established AI tools should confirm how their vendors plan to handle that transition rather than assume existing outputs are outside the rules.
A practical preparation process can start with a limited inventory and a clear ownership model:
- List customer-facing, employee-facing, and public-facing AI systems, including third-party tools.
- Identify where people could interact with an AI agent, chatbot, or automated interface without recognising it as AI.
- Review generated text, images, audio, and video to determine whether the system and use case trigger marking or disclosure duties.
- Ask vendors how they support verifiable, machine-readable content marking and what documentation they provide.
- Set publishing and approval rules for deepfakes and AI-generated public-interest content.
- Keep evidence of decisions, configurations, disclosures, and vendor information so the organisation can explain its approach.
These steps can reduce last-minute changes to websites, support flows, product interfaces, and content-production processes. They can also improve customer clarity. A concise notice at the beginning of an AI chat interaction, for instance, can set expectations before a user shares information or acts on an answer.
Enforcement and financial exposure
Enforcement will be shared among national market surveillance authorities, the AI Office for general-purpose AI related and other designated systems, and the European Data Protection Supervisor when EU institutions are providers or deployers. The AI Act provides for penalties of up to €15 million or 3% of global turnover, depending on the company, for relevant infringements.
That potential exposure makes it important not to treat transparency as a purely design or communications issue. It involves product configuration, vendor selection, content processes, and operational responsibility. However, the goal is not necessarily a complex compliance programme. For many teams, the immediate priority is to document where AI is used and put clear, proportionate notices and review steps around the use cases that Article 50 directly covers.
For businesses, transparent AI can also be a trust advantage. Users are more likely to understand the limits of an AI interaction when they know they are dealing with a system rather than a person. Clear labeling can help prevent confusion around generated media, particularly where realistic synthetic material could be mistaken for authentic content.
If your company is adding AI to customer journeys or content workflows, Scalevise's AI consultancy can help turn the Article 50 requirements into a practical implementation plan. Scalevise can assess relevant use cases, map user-facing disclosures, evaluate vendor capabilities, and prioritise the changes that reduce manual uncertainty before the 2026 deadline. Start by requesting an AI consultancy discussion.
Frequently Asked Questions
When do the EU AI transparency rules apply?
The Article 50 transparency obligations under the EU AI Act are set to apply from 2 August 2026. Certain marking obligations involving generative AI systems placed on the market before that date may have a transitional period extending to December 2026.
Will a business chatbot need to say that it uses AI?
Article 50 requires providers to inform individuals when they are interacting with an AI system. Businesses using customer-facing AI should review how their provider supports this disclosure and how it appears in the user journey.
Who must label AI-generated deepfakes?
Deployers must disclose when they publish deepfakes. They also have disclosure duties for AI-generated public-interest text. Providers have separate duties to make relevant AI-generated or manipulated content detectable through machine-readable marking.
Is the EU Code of Practice on AI-generated content mandatory?
No. The Code of Practice on Marking and Labelling AI-Generated Content is voluntary. It provides practical guidance, promotes open standards, and includes an EU icon set that can help organisations implement Article 50 transparency measures.
What are the possible penalties for non-compliance?
The AI Act provides for penalties that can reach up to €15 million or 3% of global turnover for companies, depending on the relevant infringement and company.
Conclusion
Article 50 gives businesses a clear deadline to make AI interactions and synthetic content easier to recognise. Before August 2026, the most useful work is to identify relevant AI use cases, clarify provider and deployer responsibilities, and build disclosures and labeling practices into normal product and publishing workflows. The EU's voluntary code offers a practical starting point, but each organisation remains responsible for applying the rules to its own systems and use cases.
Top comments (0)