Regfortis

EU AI Act Article 50: AI transparency and disclosure requirements

Article 50 requires businesses to tell users when they are interacting with AI and to apply appropriate disclosure practices to AI-generated synthetic content. It has applied since 2 August 2026.

Last regulatory review: 14 August 2026 · Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744

What does Article 50 require?

Article 50 of Regulation (EU) 2024/1689 imposes two distinct transparency obligations: (1) deployers of AI systems that interact with natural persons must inform those persons that they are interacting with an AI, unless obvious from context; and (2) providers and deployers of AI systems that generate synthetic content — images, audio, video, or text — must ensure that content is technically marked or accompanied by appropriate disclosure so users can identify it as AI-generated.

The two obligations are separate and can both apply to the same business. A company that uses an AI chatbot on its website (obligation 1) and also publishes AI-generated product images (obligation 2) needs to address both.

When must you disclose AI interaction?

Whenever you deploy an AI system that engages in a real-time interaction with a natural person — a chatbot, virtual assistant, AI-powered customer support agent, or automated voice system — you must inform the user that they are interacting with an AI. The disclosure must be given at the beginning of or before the interaction.

Requires disclosure

  • Website or app chatbots where users may believe they are talking to a person
  • AI-powered customer service agents answering questions
  • AI voice assistants in call centre applications
  • AI systems generating responses personalised to the user

Likely exempt ("obvious from context")

  • Clearly automated IVR phone menu systems with limited options
  • Simple keyword-triggered FAQ bots with an obvious, non-conversational interface
  • Systems already clearly labelled as automated in the product design

The "obvious from context" exception is narrow. When in doubt — particularly for conversational AI — disclosure is the safe and required approach.

Synthetic content labelling

The second part of Article 50 concerns AI-generated images, audio, video, and text that could be mistaken for human-created content. Providers of systems that generate such content must ensure it is technically marked — through watermarking, metadata, or other machine-readable signals — so that it can be identified as AI-generated. Deployers of those systems must ensure they use the technical measures the provider makes available.

For SMEs, the most common scenarios involve:

  • AI-generated product images or marketing visuals distributed publicly
  • AI-generated audio content, including voice-overs or synthetic speech in public-facing media
  • AI-generated video content published on websites, social media, or in advertising
  • Deepfake or synthetic-media content that depicts real people

Note: if the content provider (e.g. the AI tool you use) has already embedded machine-readable watermarking, your obligation may be satisfied by their technical measures — verify this with your AI tool provider.

Practical steps for SMEs

1. Audit your AI-powered customer interactions

Identify every touchpoint where your business deploys AI that talks to, responds to, or engages customers or users in real time. This includes website chat widgets, social media messaging bots, AI-powered email responders, and automated phone systems.

2. Add visible disclosure to AI interactions

At the start of each AI-mediated conversation or interaction, display a clear, plain-language disclosure. Something like "You are chatting with an automated AI assistant" is sufficient. For voice systems, a spoken disclosure at the start of the call is required.

3. Review your AI-generated content practices

If your business creates or distributes AI-generated images, audio, or video for public use, confirm whether the tool you use embeds technical labelling. If it does not, evaluate alternative tools that provide appropriate marking, or apply explicit human-readable disclosure (e.g. "Image created with AI assistance").

4. Document your disclosure approach

Record which AI systems you use that are in scope for Article 50, what disclosure method is applied to each, and the basis for any "obvious from context" exemptions you rely on. Keep this documentation up to date.

What Article 50 does NOT require

  • Disclosure for purely internal AI tools that do not interact with customers or the public
  • Disclosure when interaction with an AI is genuinely and obviously apparent from context to all reasonable users
  • Human-readable labels on every internally generated AI document
  • Prior approval from a supervisory authority before deploying an AI chatbot

Frequently asked questions

Does my website chatbot need an "AI disclosure" notice?

Yes, if the chatbot interacts with natural persons (your website visitors or customers) and is not obviously non-human from context. The disclosure should be made at the beginning of or before the interaction. A visible label such as "You are chatting with an AI assistant" placed at the start of the conversation satisfies this requirement.

What does "obvious from context" mean in Article 50?

The Act allows an exception to the disclosure requirement when "obvious from context." This means that if a reasonable user would clearly understand they are interacting with an AI — for example, an automated phone menu or a text-based ordering system with clearly non-human responses — disclosure is not required. However, this exception should be applied narrowly. If there is any realistic possibility that a user might believe they are talking to a person, disclosure is required.

Does Article 50 apply to AI that generates internal business documents?

Article 50 focuses on AI that interacts with or generates content visible to natural persons in contexts where disclosure matters — particularly consumer-facing interactions and publicly distributed AI-generated content. Internal document generation tools used purely within a business typically have less direct obligation under Article 50, though the content they produce may still need labelling if distributed externally.

Do we need to label every piece of AI-generated text?

Article 50 applies to synthetic content that could mislead people about its origin. It covers AI-generated images, audio, video, and text — but contains proportionality elements. Providers of systems generating synthetic content must ensure the content is technically marked (machine-readable watermarking or metadata). Deployers must ensure appropriate disclosure practices are in place. Not every internal AI-assisted document needs a public disclosure label, but AI-generated content distributed to the public — particularly that could be mistaken for human-created — should carry appropriate signals.

Does Article 50 apply to AI-generated emails sent to customers?

This is an area where legal interpretation is still developing. The obligation primarily concerns AI systems that interact with natural persons in real time and synthetic content that could mislead as to its origin. Personalised AI-generated marketing emails sent at scale represent an area where disclosure best practice is advisable even if the precise legal boundary is not yet fully settled by supervisory guidance.

Check your Article 50 compliance

The Regfortis assessment covers Article 50 transparency obligations alongside all other current EU AI Act requirements.

Start free assessment

Article 4: AI literacy →

Official source

Informational only — not legal advice.