What Article 50 Asks a Buyer

Transparency has applied since August 2, 2026. The December 2, 2026 date is a marking transition for systems already shipping, not a delay of the whole Act.

Edited by Emilio Santos. October 1, 2026. Updated October 24, 2026.

Who this is for

A buyer whose employees or customers in the EU will see an AI feature, including a feature switched on inside software you already pay for.

On this page

1. The dates
2. The ask
3. What Article 50 covers
4. How to run the review, step by step
5. A worked example
6. Why it is worth getting right
7. Before you sign off
8. FAQ

This is an operator checklist, not a legal opinion and not a finding that your use is covered. The European Commission’s AI Act Service Desk says Article 50 transparency rules started to apply on August 2, 2026. It also says that for certain generative systems already placed on the market before that date, the Article 50(2) machine-readable marking duty has a transition that ends December 2, 2026. The Council’s timeline records the high-risk delay separately: December 2, 2027 and August 2, 2028. Do not let a vendor use the high-risk date to answer a transparency question.

The dates

The ask

• Is a person told they are interacting with an AI system, where the Act requires it? Ask for the screen, not the policy.
• If the system generates synthetic audio, image, video, or text, what machine-readable mark do you apply, and as of which date?
• Was this system on the market before August 2, 2026? If yes, the vendor should be able to say whether they are inside the December 2, 2026 transition.
• Which article and annex are you relying on when you say “not until 2027”?

What Article 50 covers

Article 50 of the AI Act, Regulation (EU) 2024/1689, sets transparency duties for certain AI systems. In plain terms:

• Interacting with AI. Providers must design systems that interact directly with people so that people are told they are interacting with an AI system, unless that is obvious from the context.
• Marking synthetic content. Providers of systems that generate synthetic audio, image, video, or text must mark the outputs in a machine-readable format so they can be detected as artificially generated or manipulated.
• Emotion recognition and biometric categorisation. Deployers of these systems must inform the people exposed to them.
• Deepfakes and public-interest text. Deployers that generate or manipulate deepfake image, audio, or video must disclose that it is artificial. Deployers that publish AI-generated text to inform the public on matters of public interest must disclose it, with exceptions such as where the text has had human review and someone holds editorial responsibility.

Note the split between provider duties and deployer duties. If you buy a product and use it, you may be a deployer. The vendor is usually the provider. Your contract should say who does what.

How to run the review, step by step

1. List AI features that reach people in the EU. Chat assistants, generated images or audio in marketing, AI-written customer messages, and features inside software you already own.
2. Tag each one by Article 50 category. Interaction, synthetic content, emotion recognition, deepfake, or public-interest text. Some will fit none.
3. Ask the vendor the questions above. Get the answers in writing, with dates.
4. Assign your deployer duties. For deepfakes and public-interest text, the disclosure may be your job, not the vendor’s.
5. Record the decision. Keep a dated note of what you reviewed and why. Have counsel confirm where the answer is not obvious.

A worked example

Illustrative case. A company uses a customer service chat assistant on its EU website and an image generator for social media ads. For the chat assistant, the buyer asks the vendor to show the screen where customers are told they are talking to an AI system. For the image tool, the buyer asks what machine-readable mark is applied to generated images, and whether the tool was on the market before August 2, 2026, which would put it in the transition ending December 2, 2026. If any ad image is a realistic depiction of a real person or event, the company, as deployer, plans its own disclosure.

Why it is worth getting right

The AI Act sets fines for breaches of the transparency obligations of up to 15 million euros or 3% of total worldwide annual turnover, whichever is higher, with lower caps for small and medium-sized companies. That is a reason to keep a record of the review, not a reason to panic. The timeline for everything else in the Act is in What the EU AI Act still applies.

Before you sign off

• AI features that reach people in the EU are listed.
• Each is tagged by Article 50 category, or marked as outside it.
• Vendor answers on disclosure screens and machine-readable marks are in writing, with dates.
• Deployer duties, such as deepfake disclosure, have an owner on your side.
• Counsel has reviewed the cases that are not obvious.

FAQ

We are a US company. Does this matter?

It matters if the vendor, the user, or the customer is in the EU, or if the contract pushes EU duties onto you. The checklist is the question list. Counsel decides whether you are a provider or a deployer.

Where is the law?

Regulation (EU) 2024/1689 on EUR-Lex, read with the Commission timeline that includes later amendments. Do not brief from a blog’s date if it disagrees with those two.

Sources

• European Commission AI Act Service Desk timeline
• Council of the EU, AI Act timeline
• Regulation (EU) 2024/1689 on EUR-Lex
• Regulation (EU) 2024/1689, the AI Act, on EUR-Lex: Article 50 transparency obligations and Article 99 penalties

Related

• The EU AI Act did not slip. The high-risk part did.
• When compliance is the product