SUMMARY:

• On 20th July, the European Commission published the final version of the guidelines on AI Act transparency obligations for providers and deployers of certain AI systems, ahead of their application from 2 August 2026.‍
• The final guidelines largely maintain the draft approach but provide more detailed and operational guidance on the definition of deep fakes, expanding the scope to cover digital replicas, distinctive personal characteristics, and plausibly existing subjects, while introducing a more contextual assessment of whether content would falsely appear authentic or truthful. They also refine the practical application of the disclosure obligation, particularly for artistic/satirical content, distribution contexts, and deployer responsibilities (see below for more information).
• The Commission explains that compliance may be demonstrated through adherence to the Code of Practice on Transparency of AI-generated content.

MORE INFORMATION:

Some of the key changes introduced by the guidelines (compared to the draft version):

1. Definition of deep fakes (section 6.1.1)

• "Resemblance" (i) is broadened: can include: recognisable elements; characteristic features; distinctive features.
• “Existing” (ii) is expanded: the final guidelines changes the threshold from: subjects that “can exist or could have existed” to: subjects that “exist, can plausibly exist or could have plausibly existed”.
• “Persons” (iii) is clarified: can include digital replicas of real people; realistic AI-generated human avatars; personal characteristics or expressions (such as: image; voice; behaviour; performances etc).

2. Assessment of whether content “would falsely appear authentic or truthful” (section 6.1.1 iv):

• The draft focused mainly on: whether people could be deceived; audience composition; whether children, elderly persons or less digitally literate users might be misled. It stated that the deployer’s intention to deceive was irrelevant.
• The final version introduces a more structured objective assessment, which should include:
  
  • The content itself, including: resemblance;substantive message.
  • Deployment context, including: where it appears; how it is presented;surrounding circumstances.
  • Audience expectations, including: intended audience; reasonably foreseeable audience; whether the audience expects authenticity.
  • Audience composition: not "average" person but taking into account potentiel perceptions by children, the elderly, or other groups of persons with lower digital and AI literacy or general knowledge

3. Minor AI editing and enhancement (section 6.1.1 iv §114)

• The draft already stated that not every AI-assisted modification creates a deep fake—>ex:  colour correction; noise reduction; compression; accessibility improvements; background adjustments.
• The final keeps this principle but places it within the broader contextual assessment, meaning: AI modification becomes relevant where it affects perceived authenticity or truthfulness.

4. Disclosure obligation under Article 50(4) (Section 6.1.2)

• The draft stated that deployers must disclose that deep fake content has been artificially generated or manipulated.

• The final version adds more practical clarification around:
  
  • first exposure;
  • presentation context;
  • visibility of the disclosure;
  • interaction with Article 50(5)'s clear and distinguishable information requirement.

• The final also links the disclosure obligation more closely to the role of deployers in content production and distribution chains.
• The draft adds that deployers may adhere to the code of practice to demonstrate compliance.

5. Artistic, creative, satirical, fictional and analogous works

• The draft recognised that artistic and creative freedom requires a different approach.
• The key refinement is that the exemption does not mean no transparency is needed. Instead:
  
  • disclosure can be adapted;
  • transparency should not undermine artistic expression;
  • the audience should still understand the artificial nature of the content where required.

• From our reading (FESI views only), marketing materials may potentially benefit from this flexibility where they involve genuine creative, fictional or expressive elements (for example, fictional worlds, stylised narratives or clearly imaginative scenarios), but commercial intent alone does not remove the disclosure obligation where content could reasonably be perceived as authentic or truthful.

From our perspective, the final guidelines represent a somehow positive evolution by focusing more clearly on authenticity, context and user perception rather than the mere use of AI, which aligns with the direction advocated by FESI. However, they do not fully resolve the uncertainty surrounding the treatment of commercial marketing content, nor do they provide the sector-specific guidance that FESI had requested.