What changed on 2 August 2026?

Article 50 moved from preparation to active compliance. The rules now cover four practical situations: direct interaction with an AI system; machine-readable marking of synthetic outputs by providers; notice for emotion recognition or biometric categorisation; and disclosure for deepfakes and certain AI-generated public-interest text.

Article 50 compliance checklist

  1. Inventory the use cases. Record customer-facing AI, synthetic-content generation, emotion recognition, biometric categorisation and deepfake workflows.
  2. Confirm your role. Document whether you are acting as provider, deployer or both for each system and output.
  3. Map the applicable paragraph. Record whether Article 50(1), (2), (3) or (4) applies and any exception relied on.
  4. Implement the notice. Make disclosures clear, accessible and presented no later than the first interaction or exposure.
  5. Verify machine-readable marking. Providers of covered generative systems should test that synthetic outputs are marked and detectable.
  6. Control public-interest publishing. Keep evidence of human review, editorial control and the responsible person or organisation.
  7. Test and retain evidence. Save screenshots, test results, supplier specifications, decisions, owners and review dates.

What must providers do?

Providers of systems designed to interact directly with people must build in a disclosure unless the interaction is obvious. Providers of covered systems that generate synthetic audio, image, video or text must also support machine-readable marking and detection, subject to the Act's limits and exceptions.

What must deployers do?

Deployers must inform people exposed to emotion recognition or biometric categorisation. They must also disclose covered deepfakes and AI-generated or manipulated public-interest text, unless the text has undergone human review or editorial control and editorial responsibility is held by a person or organisation.

What evidence should you keep?

A defensible evidence pack should include the system inventory, role assessment, Article 50 mapping, notice wording, deployment screenshots, accessibility checks, marking or detection test results, supplier documentation, editorial-review records, exceptions, approvals, owners and the next review date.

Use the post-deadline UK business evidence checklist for a broader record set beyond Article 50.

Frequently asked questions

Is Article 50 of the EU AI Act now in force?

Yes. Article 50 transparency obligations have applied since 2 August 2026. A limited transition to 2 December 2026 applies only to the Article 50(2) marking obligation for certain generative AI systems placed on the market before 2 August 2026.

Who must tell people they are interacting with AI?

Providers of AI systems intended to interact directly with people must design the system so the person is informed, unless the AI interaction is obvious in context. Deployers should verify that the disclosure is present and works in the deployed experience.

Does every piece of AI-assisted text need an AI label?

No. The deployer disclosure duty for public-interest text does not apply where the content underwent human review or editorial control and a person or organisation holds editorial responsibility. Providers may still have machine-readable marking duties for covered synthetic outputs.

What evidence should a business retain?

Keep an inventory of affected systems and content workflows, role and scope decisions, screenshots or test records of notices, supplier marking specifications, editorial-review records, exceptions relied on, owners, approval dates and periodic review evidence.

Does Article 50 apply to UK businesses?

It can. The AI Act can reach a UK provider placing a system on the EU market and organisations outside the EU where an AI system's output is used in the EU. Scope should be assessed system by system.

Related pages

Official sources

Last updated 4 August 2026. This is general information, not legal advice.