Does Article 50 apply to a business outside the EU?

It applies where the AI system is placed on the EU market or its output is used in the EU. A UK or US business with EU customers who see its chatbot or AI-generated content should assume it is in scope for those interactions.

Do I need to label every image I make with an AI tool?

The deployer duty is about deepfakes: content that depicts real people, places, objects or events in a way that would falsely appear authentic. An obviously illustrative graphic is lower risk, but labelling AI-generated visuals as a matter of policy is simpler than judging each one.

Our chatbot clearly looks like a bot. Do we still need a disclosure?

The Act excuses disclosure only where the AI nature is obvious to a reasonably well-informed, observant person. A one-line statement costs nothing and removes the argument, so include it.

We use AI to draft blog posts that a human edits. Is disclosure required?

Where the text informs the public on matters of public interest, disclosure is not required if the content has undergone human review and a natural or legal person holds editorial responsibility. Keep a record of who reviewed it. For ordinary marketing copy the public-interest test is unlikely to be met, but a light disclosure does no harm.

Who is responsible for watermarking, us or the tool vendor?

The machine-readable marking duty in Article 50(2) sits with the provider of the generative system. Your duty as a deployer is the visible disclosure to people who see the content, plus checking that your vendor meets its own obligation.

Guides · Updated 2026-09-06

EU AI Act Article 50: Transparency Rules for Small Business

Article 50 of the EU AI Act sets out transparency obligations for AI systems that interact with people or generate content. Those obligations became applicable on 2 August 2026, and unlike the high-risk provisions they were not pushed back by the Digital Omnibus. If your business operates a chatbot, publishes AI-generated images or audio, or uses AI to write text for the public, this guide explains what you now have to do and how to document it.

The free generator produces an AI Transparency Notice alongside your policy so the Article 50 disclosures are written down and ready to publish.Start the free generator

What Article 50 requires

Article 50 is deliberately technology-neutral and applies regardless of company size. It creates four practical duties. First, when people interact with an AI system such as a chatbot or voice assistant, they must be informed that they are dealing with AI, unless that is obvious from the circumstances to a reasonably informed person. Second, providers of systems that generate synthetic image, audio, video or text must ensure the output is marked in a machine-readable format so it can be detected as artificially generated. Third, deployers who use AI to produce or manipulate deepfakes, meaning content that resembles real people, places or events and would falsely appear authentic, must disclose that the content was artificially generated or manipulated. Fourth, deployers who publish AI-generated text to inform the public on matters of public interest must disclose that it was AI-generated, unless the text has been through human review and a person or organisation holds editorial responsibility for it.

The Act also requires deployers of emotion recognition or biometric categorisation systems to inform the people exposed to them, with limited exceptions. For most small businesses that fourth category will not apply, but it is worth checking if you use any HR or customer analytics tool that claims to read mood or sentiment from faces or voices.

What changed with the Digital Omnibus, and what did not

The Digital Omnibus package moved the application dates for high-risk AI system obligations to 2027 and 2028. It did not move Article 50. The transparency duties for chatbots, deepfakes and AI-generated public-interest text have applied since 2 August 2026, and businesses should treat them as live obligations today.

There is one carve-out. The Article 50(2) machine-readable marking requirement, which falls on providers of generative systems, applies from 2 December 2026 for systems that were already on the market before the general application date. If you build or resell a generative tool, that is the date to plan around for watermarking and metadata. If you merely use someone else's tool, the marking duty is theirs; your duty is disclosure to the people who see the output.

Separately, the Article 4 AI literacy obligation has applied since 2 February 2025. It requires providers and deployers to ensure that staff and others operating AI systems on their behalf have a sufficient level of AI literacy. That duty runs in parallel with Article 50 and is covered in its own guide.

What a small business must actually do

Start by listing every place where your business either puts an AI system in front of a person or publishes AI-generated content. Typical entries for a small company are a website chatbot, an automated phone or WhatsApp assistant, AI-generated product images or social media visuals, voice-over or music produced by AI for videos, and blog or news-style content written with AI assistance.

For each entry, decide which of the four duties applies and record the disclosure you will make. The Act does not prescribe wording, so plain language works: a chatbot can open with a sentence saying it is an automated assistant, an image can carry a caption or visible label, and an article can carry a note that AI tools were used in drafting. Where a human edited and takes responsibility for a text, note who that was and keep the editing record, because that is what removes the disclosure requirement for public-interest text.

Add a clear AI disclosure to every chatbot, voice bot and automated messaging flow at the start of the conversation.
Label AI-generated or AI-manipulated images, audio and video that depict real people, places or events. Do not rely on invisible metadata alone for content your audience will see.
For AI-assisted articles on matters of public interest, either disclose the AI use or document human review and named editorial responsibility.
Ask your AI tool suppliers whether their outputs carry machine-readable markings and keep their answers on file.
Record all of the above in a short AI Transparency Notice and link it to your acceptable use policy.

What an AI Transparency Notice should contain

An AI Transparency Notice is a short public-facing document, usually one page, that explains how your business uses AI in ways that affect the people reading it. It is not required by name in the Act, but it is the simplest way to show that the Article 50 duties have been thought through, and it gives customers a single place to look.

Include a statement of which customer-facing interactions are automated, how you label AI-generated content, whether AI is used in any decisions that affect individuals and how a person can request human review, a contact point for questions, and the date of the last update. Keep the internal version slightly longer, with the inventory of systems that sit behind each statement, so that the public notice can be defended if anyone asks.

Penalties and proportionality

Fines for transparency breaches can reach 15 million euros or 3 percent of worldwide annual turnover, whichever is higher, and the Act directs regulators to consider the size of the business and the interests of SMEs when setting penalties. In practice, a small business with a written notice, labelled content and a record of supplier checks is in a very different position from one that has never considered the question. The point of documenting is not to eliminate risk but to demonstrate that a reasonable effort was made.

Exceptions exist for law enforcement uses authorised by law and, for artistic, satirical or creative works, the disclosure need only be made in a way that does not hamper the display or enjoyment of the work. Those exceptions are narrow and should not be relied on for ordinary commercial content.

Frequently asked questions

Does Article 50 apply to a business outside the EU?: It applies where the AI system is placed on the EU market or its output is used in the EU. A UK or US business with EU customers who see its chatbot or AI-generated content should assume it is in scope for those interactions.
Do I need to label every image I make with an AI tool?: The deployer duty is about deepfakes: content that depicts real people, places, objects or events in a way that would falsely appear authentic. An obviously illustrative graphic is lower risk, but labelling AI-generated visuals as a matter of policy is simpler than judging each one.
Our chatbot clearly looks like a bot. Do we still need a disclosure?: The Act excuses disclosure only where the AI nature is obvious to a reasonably well-informed, observant person. A one-line statement costs nothing and removes the argument, so include it.
We use AI to draft blog posts that a human edits. Is disclosure required?: Where the text informs the public on matters of public interest, disclosure is not required if the content has undergone human review and a natural or legal person holds editorial responsibility. Keep a record of who reviewed it. For ordinary marketing copy the public-interest test is unlikely to be met, but a light disclosure does no harm.
Who is responsible for watermarking, us or the tool vendor?: The machine-readable marking duty in Article 50(2) sits with the provider of the generative system. Your duty as a deployer is the visible disclosure to people who see the content, plus checking that your vendor meets its own obligation.

Generate your own in about four minutes

The free generator produces an AI Transparency Notice alongside your policy so the Article 50 disclosures are written down and ready to publish.

Generate my policy free

Related guides

This guide is general information, not legal advice.