EINBLICK / AI Act news / AI transparency / AI content / Deepfake / EU regulation

EU AI Act transparency enforcement started on 2 August 2026: what changed

Chatbots, deepfakes and AI content: the EU transparency rules now being enforced and the checks AI product owners should run.

On 2 August 2026, the European Commission AI Office and national authorities started enforcing applicable provisions of the EU AI Act. Transparency duties for certain AI systems began applying on the same date. This is a practical change for businesses operating chatbots, AI avatars, and image, video, audio or public-text generation.

Which scenarios are affected

  • Interactive AI. A person should understand that they are interacting with AI where that is not obvious from context.
  • Synthetic content. Providers need to support detection of AI-generated or manipulated output, including machine-readable marking where the rule applies.
  • Deepfakes. Generated or altered image, audio and video content may require clear disclosure.
  • Public-interest text. Certain AI-generated material published to inform the public is subject to disclosure, taking account of the exceptions in the rules.

The Commission has also published Article 50 guidance. Businesses should check more than the visible label: identify who is the provider or deployer, how the content is published and whether a mark survives export to a social or advertising platform.

What product owners should do now

  1. list chatbots, voice agents, AI avatars and generative features;
  2. identify the company's role and the relevant transparency scenario;
  3. place the notice at the correct point in the user journey;
  4. test metadata and machine-readable marking after download and publishing;
  5. update product requirements, vendor contracts and content workflows;
  6. train marketing, support and sales teams on labelling.

See the detailed EU AI Act business checklist for an implementation plan. CrystalCore can build requirements into the interface, data flows and generation pipeline through its AI platform service.

What the change does not mean

It does not make every AI system high-risk or require certification for every chatbot. Classification depends on intended purpose and context. The high-risk calendar was also amended in July 2026, with separate later dates for Annex III and Annex I systems.

Sources

Updated on 10 October 2026. This news item is not legal advice.

Anfrage senden