Deepfakes and the New Disclosure Requirement: What Users (Legal Entities and Individuals) of AI Systems Need to Know from August 2026

What Is Considered a Deepfake Under the Law?

The legal definition of a deepfake is narrower than many people assume. According to Article 3(60) of the AI Act, a deepfake is an AI-generated or AI-manipulated image, audio recording, or video that resembles a real person, object, place, entity, or event and is capable of being falsely perceived as authentic or truthful.

To clarify this definition, the European Commission published its final Guidelines on Article 50 of the AI Act on 20 July 2026. In addition, a Code of Practice on Transparency of AI-Generated Content, published on 10 June 2026, provides practical guidance for businesses on implementing the transparency requirements.

The classification of content as a deepfake depends primarily on four criteria:

1. Recognizable resemblance: The content must closely resemble a person, place, object, or event.
2. Real or plausible existence: The depicted subject must exist, could plausibly exist, or could plausibly have existed.
3. Broad scope of application: The rules apply not only to people but also to places, objects, and events, including digital replicas of real individuals and highly realistic AI-generated avatars.
4. Potential to mislead: The content must be capable of leading its intended audience to believe that it is genuine or truthful.

Importantly, an intention to deceive is not required. Nor is the assessment based on the perception of an average person. Instead, the relevant benchmark is the actual target audience, which may include older individuals, children, or users with limited media literacy.

By contrast, content that is obviously unrealistic, such as people flying without assistance or talking animals, generally does not fall within the legal definition of a deepfake.

Practical Example

A photorealistic AI-generated avatar appears in a promotional video advertising a real smartphone model. The avatar’s facial expressions, speech, and movements appear convincingly human, even though the character itself is entirely fictional.

Because the avatar closely resembles a real person and is capable of creating the impression of authenticity, it qualifies as a deepfake subject to the AI Act’s transparency requirements, even though the depicted individual does not actually exist.

If the video also features a real product protected by trademark rights, businesses should assess not only the AI Act’s disclosure obligations but also potential issues under trademark law and unfair competition law.

How Must Deepfakes Be Properly Disclosed?

The disclosure must be clear, unambiguous, and easily understandable. Users should be able to recognize immediately that the content has been generated or manipulated by AI, without requiring technical tools or additional investigation.

According to the European Commission’s Guidelines, it is not sufficient to provide the disclosure only:

• in metadata,
• in general terms and conditions, or
• in the website footer.

Commonly accepted wording includes statements such as:

• “AI-generated”
• “This content was created or modified using artificial intelligence.”

For images, the disclosure should appear directly within the image itself, for example by displaying a clearly visible icon or notice in one of the corners.

For pre-recorded videos, a notice at the beginning of the video is generally recommended. For live streams, the disclosure should remain visible throughout the broadcast.

The European Commission has also introduced standardized icons for AI-generated content. Their use, however, remains voluntary.

Are There Any Exceptions?

The disclosure obligation does not apply where the use of deepfakes is legally authorized for the detection, prevention, investigation, or prosecution of criminal offences.

Special rules also apply to clearly artistic, creative, satirical, or fictional works. In such cases, the disclosure may be presented in a way that does not interfere with the audience’s enjoyment of the work. For example, a notice in the opening or closing credits of a film may be sufficient.

Purely technical image enhancements—such as exposure correction, noise reduction, or routine retouching—are generally not considered deepfakes.

However, if AI editing creates a new, highly realistic depiction that may be perceived as authentic, the disclosure requirement will apply.

Whether a particular piece of content qualifies as a deepfake must always be assessed on a case-by-case basis. Where any uncertainty exists, a transparent approach to disclosure is generally preferable to a restrictive one.

What Are the Risks of Failing to Comply with the Disclosure Requirement?

Failure to comply with the transparency obligations under the AI Act may result in significant administrative fines.

In addition, misleading AI-generated advertising may expose businesses to cease-and-desist letters and legal claims brought by competitors or consumer protection associations under unfair competition law. This may be the case, for example, where AI-generated images present a company, product, or professional practice in an unrealistically favorable manner.

If real trademarks or branded products are featured in undisclosed deepfake content, businesses may also face claims under trademark law as well as reputational damage.

Furthermore, major social media platforms have introduced their own transparency requirements for AI-generated content. Failure to comply with these platform rules may lead to reduced visibility, content removal, or other account restrictions.

When Does the Disclosure Requirement Apply?

The mandatory disclosure requirement for deepfakes applies from 2 August 2026. Although the proposed AI Omnibus Package is expected to postpone certain implementation deadlines under the AI Act, the current effective date for the deepfake transparency obligation remains unchanged.

Our Recommendation for Businesses and Other Users

Companies and individuals using AI to create advertising materials, social media content, or training materials should act now.

In particular, we recommend that you:

• systematically assess which of your AI-generated content qualifies as a deepfake;
• implement clear and consistent disclosures for AI-generated images, videos, and avatars;
• review trademark, copyright, and unfair competition law implications, especially where real products, brands, or individuals are depicted; and
• ensure that your marketing and content teams are fully trained before the new rules take effect.

The interaction between AI law, trademark law, copyright law, and unfair competition law is complex, and non-compliance may expose businesses to substantial legal liability.

As a law firm specializing in intellectual property law, SCHNELLER LEGAL advises companies on how to implement legally compliant AI-generated content, review existing marketing campaigns, and minimize the risk of cease-and-desist letters, litigation, and administrative fines.

Feel free to contact us for tailored legal advice on the use of AI-generated content in your business.