Guide · EU AI Act · Article 50

EU AI Act Article 50 disclosure checklist

By Mayank Raj · Kingdom of Kumar · Updated 4 November 2026

Article 50 of the EU AI Act, the transparency rules for chatbots, AI-generated content and deepfakes, has applied since 2 August 2026. The one later date is 2 December 2026, and it covers only the machine-readable marking duty in Article 50(2) for AI systems that were already on the market before 2 August. This page lists what each role has to do and what evidence to keep. It is not legal advice.

Not legal advice. This is a plain-language checklist built from the official texts linked below. Whether and how Article 50 applies to your product is a legal question for you and your lawyer.

EU AI Act Disclosure Pack, $497 setup →

Tooling for banners, JSON-LD labels and an audit checklist. It does not scan your site and does not replace a lawyer.

The dates that matter

What the Digital Omnibus did and did not do: it moved Annex III stand-alone high-risk obligations to 2 December 2027 and Annex I product-embedded high-risk obligations to 2 August 2028, per Hunton’s summary. It did not defer Article 50. Recital 38 adds a four-month transitional period for Art. 50(2) providers already on the market before 2 August 2026, which is the source of the 2 December date. The Commission also confirms that content generated before 2 August 2026 does not have to be labelled retroactively, though it encourages doing so where possible.

What Article 50 requires

Read the text on the AI Act Service Desk. In summary:

• 50(1), providers: people must be told they are interacting with an AI system, unless that is obvious.
• 50(2), providers of generative systems: outputs must be marked in a machine-readable format, with exceptions for assistive standard editing and for systems that do not substantially alter input.
• 50(3), deployers of emotion recognition or biometric categorisation: inform the people exposed to it.
• 50(4), deployers: disclose deepfakes (a lighter duty for evidently artistic, satirical or fictional work), and disclose AI-generated or manipulated text published to inform the public on matters of public interest, unless there is human review or editorial control and someone holds editorial responsibility.
• 50(5), everyone: information must be clear and distinguishable, given at the latest at first interaction or exposure, and meet accessibility requirements.

Checklist for providers

1. List every AI system you place on the market and mark which ones interact directly with people, which generate synthetic audio, image, video or text, and which do both.
2. For direct interaction, add a disclosure at the first message or first exposure, unless AI use is obvious from context (50(1), 50(5)).
3. For generative outputs, implement machine-readable marking and a way to detect it (50(2)). Note the date: if the system was on the market before 2 August 2026, you have until 2 December 2026.
4. Record whether an exception applies (assistive standard editing, no substantial alteration of input) and why.
5. Read the Commission Guidelines and decide whether to adhere to the Code of Practice.

Checklist for deployers

1. List where you use AI systems with people, and where you publish AI-generated or manipulated images, audio, video or text.
2. For deepfakes, add a visible disclosure (50(4)). Use the lighter form only where the work is evidently artistic, satirical or fictional.
3. For AI-generated text published to inform the public on matters of public interest, either disclose it or document the human review or editorial control and who holds editorial responsibility.
4. If you use emotion recognition or biometric categorisation, inform the people exposed (50(3)).
5. Place disclosures at first exposure, in plain words, and test them with screen readers and on mobile (50(5)).
6. Consider the Commission’s optional icons for labelling AI content.

If you are an agency, you may be a deployer for your own output and a provider for a product you build and offer under your own name. Settle the role per client, with your lawyer and in writing, before you build.

Evidence to keep

• An inventory of AI systems and features, with your role (provider or deployer) and the reasoning.
• Screenshots or recordings of each disclosure as shown to users, with dates.
• Technical notes on how marking works, and test results showing it can be detected.
• Records of any exception relied on and who approved it.
• For published text, the human review or editorial control record and the named editor.
• Which version of the Guidelines and Code of Practice you read, and whether you adhere to the Code. Adherence is not conclusive evidence of compliance.
• A change log, so you can show what was live on a given day.

Fines and who enforces

Under Article 99(4), breaches of operator obligations including Article 50 can be fined up to EUR 15,000,000 or 3% of total worldwide annual turnover, whichever is higher. For SMEs, Article 99(6) applies the lower of the two. Enforcement sits with national market surveillance authorities, the AI Office for systems under its supervision, and the EDPS for EU institutions (Commission guidelines page).

Questions

Related guides

• A 2027 AI plan for agencies
• Cut AI costs before your 2027 budget
• AI agent for holiday customer support