What Automakers, Suppliers, and Agencies Need to Know Starting in August 2026
The use of generative AI has fundamentally transformed the creation of marketing and communications content. Today, promotional videos, vehicle images, social media campaigns, and virtual brand ambassadors can be generated entirely or partially by AI within a matter of minutes. Especially in the automotive industry, where emotional imagery, product presentations, and digital customer interactions play a central role, the use of such technologies is steadily increasing.
The transparency requirements of the European AI Act take effect on August 2, 2026, and require companies to label certain AI-generated content. It should be noted that these obligations apply only to professional activities, not to purely personal ones (Art. 3 (4) of the AI Regulation), and that a transition period applies to the machine-readable labeling of systems already on the market (see section 2 below). Practical implementation raises numerous questions. In particular, it has not yet been conclusively clarified when an AI-generated (vehicle) image is to be classified as a “deepfake” within the meaning of the regulation, nor what form of labeling will be sufficient.
The following article provides an overview of the new requirements and outlines how companies can successfully approach implementation.
1. When is labeling required?
The labeling requirements stem from Article 50 of the AI Regulation. The provision distinguishes between two parties subject to these requirements. Operators (“deployers”) must, pursuant to Article 50 (4) of the AI Regulation, disclose that published image, audio, or video content has been artificially generated or manipulated (deepfake disclosure). In addition, providers of generative AI systems are required under Article 50 (2) of the AI Regulation to technically label the output - in a machine-readable format - as artificially generated.
The subject of disclosure obligation under Article 50(4) of the AI Regulation is generally not the provider of the AI system, but rather the company that actually uses or publishes the content. An automaker that publishes an AI-generated promotional video on its website or on social media will therefore typically be considered the operator.
2. What is a deepfake?
In practice, the term “deepfake” is often reduced to manipulated images of people. The AI Regulation, however, takes a significantly broader approach. Article 3 (60) of the AI Regulation requires that the content “resembles real persons, objects, places, facilities, or events and would falsely appear to a person to be genuine or truthful.”
This covers content that
- has been artificially generated or altered,
- contains realistic-looking depictions, and
- is likely to give the viewer the impression that it consists of authentic footage.
Contrary to common belief, therefore, it is not necessary for people to be depicted. Vehicles, buildings, production facilities, trade show booths, or entire landscapes may also be covered. According to the Commission’s draft guidelines, however, depictions that are obviously unrealistic or physically impossible (e.g., surreal or recognizably fantastical motifs) do not fall under the definition of “deepfakes.”
Automotive companies, in particular, are already using numerous applications today that could potentially fall under the transparency requirements.
Examples include:
- entirely AI-generated commercials,
- AI-generated vehicle images for marketing purposes,
- virtual test drive scenarios,
- lifestyle campaigns featuring AI-generated people,
- AI-generated trade show and event visualizations,
- virtual brand ambassadors,
- synthetic voice recordings,
- AI-generated product presentations on websites or in apps.
It is particularly important to note that the disclosure requirement does not apply only in cases of intent to deceive; what matters is whether the content is capable of being mistaken for the real thing. With regard to deepfake disclosure under Article 50 (4) of the AI Regulation, the regulation does not provide for an explicit de minimis threshold; even short video clips or individual images may be subject to the labeling requirement. By contrast, an exception applies to the machine-readable provider identification under Article 50 (2) of the AI Regulation if the system performs only a “supporting function for standard processing” or “does not significantly alter” the input data.
3. Are AI-generated vehicle images also deepfakes?
This is currently one of the most important unresolved legal questions. When it comes to AI-generated images of people, classification is generally unproblematic. The situation becomes more difficult with product depictions.
For example: A manufacturer creates an advertising image of a vehicle entirely using generative AI. The vehicle model actually exists. However, the image depicted was never photographed. The AI Regulation has not yet explicitly addressed whether such product images should also be considered deepfakes.
Arguments in favor of a labeling requirement include:
- Vehicles are expressly covered as “objects” under Article 3 (60) of the AI Regulation.
- The viewer might assume that the image is a genuine photograph.
- The transparency requirement is specifically intended to disclose the origin of the content.
Arguments against a labeling requirement include:
- Consumers are often aware that marketing images and vehicle configurators are based on renderings.
- Depictions of vehicles typically do not pose a particular risk of infringing on personality rights.
- Product visualizations have been a standard marketing tool for decades.
To date, there is no established regulatory or judicial precedent on this matter. Until further clarification is provided by the Commission or national regulatory authorities, a more “cautious” approach is therefore recommended, if one wishes to avoid any complaints. However, current practice already shows that there is certainly room for interpretation here.
4. What applies to traditional CGI renderings and configurators?
Not every computer-generated image automatically falls under Article 50 of the AI Regulation. In the case of traditional CAD renderings or CGI images, such as those used for many years in vehicle configurators, there are strong arguments against classifying them as deepfakes. The key difference is that generative AI is generally not used here; instead, conventional visualization technologies are employed. Furthermore, users of a vehicle configurator typically do not expect a real photograph of a specific vehicle, but rather an abstracted product representation. However, a clear distinction will not always be possible in practice, especially when modern generative AI systems are integrated into existing rendering processes.
5. What about chatbots and virtual assistants?
In addition to deepfakes, Article 50 of the AI Regulation contains further transparency requirements. According to Article 50 (1) of the AI Regulation, the provider must “design and develop” an AI system intended for direct interaction with natural persons in such a way that users recognize they are interacting with an AI (compliance-by-design). This obligation does not apply if this is “obvious from the perspective of a reasonably informed, observant, and circumspect natural person.”
This applies, for example, to AI-powered sales assistants, customer service chatbots, virtual vehicle advisors, voice assistants on websites, and AI-based support features in apps. For manufacturers and dealers, this means they should ensure that the AI is clearly identifiable within their own user interfaces so that users can recognize from the very beginning of the interaction that they are communicating with an AI.
6. What should be considered when using AI in the head unit?
The number of use cases is growing rapidly: AI voice assistants for vehicle and navigation control, conversational in-car assistants based on large language models, AI-powered concierge and recommendation services, generated voice output, and AI-powered in-vehicle customer support and service functions. Legally, such interactive systems are generally classified as AI systems intended for direct interaction with natural persons and are therefore subject to the transparency requirement under Article 50 (1) of the AI Regulation. The driver or passenger must therefore be able to recognize that they are communicating with an AI and not with a human.
The same standard applies to the head unit as to chatbots on the web or in apps: The obligation primarily falls on the provider, who must “design and develop” the system in such a way that users are informed about the AI interaction (compliance-by-design). This obligation does not apply only to the extent that the use of AI is “obvious from the perspective of a reasonably informed, attentive, and discerning natural person.”
In practice, this exemption is limited to cases where the use of AI is evident; precautionary notices are advisable even to avoid legal uncertainties. Furthermore, if the system generates synthetic speech, image, or video output, the machine-readable labeling requirement under Article 50(2) of the AI Regulation - which applies to the provider - may also apply. For embedded vehicle systems that operate in a technically controlled, closed environment and whose output does not leave the product, the Code of Practice considers a single layer of labeling sufficient in this regard.
Questions are likely to arise in particular when third-party services are integrated into the vehicle, e.g., the service of an external voice control provider. The respective responsibilities will depend largely on the specific implementation and on how the service is actually used and perceived by the user.
In any case, it is recommended that automotive companies address the issue of transparency notices early in the development process, integrate them firmly into the head unit’s user experience (UX) where necessary (e.g., an audio or visual notice at the start of the interaction, labeling of generated voice outputs), and establish these requirements contractually in their relationships with suppliers of infotainment and voice systems.
7. How must the labeling be carried out?
The AI Regulation does not contain detailed design specifications. According to Article 50 (5) of the AI Regulation, the information must be provided “clearly and unambiguously no later than at the time of the first interaction or suspension” and must be accessible without barriers.
Further details are provided by the Commission’s draft guidelines as well as the accompanying Code of Practice on Transparency of AI-Generated Content, a voluntary practical guide pursuant to Article 50 (7) of the AI Regulation. It is important to note that the Code is voluntary and expressly does not constitute conclusive evidence of compliance; adherence to it merely allows for the demonstration of compliance. It is divided into two sections—one for providers (Article 50 (2)/(5)) and one for operators (Article 50 (4)/(5)).
Accordingly, the labeling must, in particular, be clearly visible, understandable, provided in a timely manner (no later than the first suspension), and impossible for the average user to overlook. Hidden disclosure in the legal notice, footer, or terms of use is generally unlikely to suffice, because the icon must be visible without user interaction and without requiring special attention.
The Official EU Icon
As part of the Code, the AI Office provides a freely usable EU icon that represents the simplest, most uniform way to comply with the disclosure requirement. There are three variants - “AI GENERATED” for content fully generated by AI, “AI MODIFIED” for partially manipulated content, and a basic icon—which can be used without an attribution requirement. The Code specifies the uppercase abbreviation “AI” as the main element; if English is not permitted under national law, the local language may be used. Empirical user tests have shown that variants with text (“modified”/“generated”) are significantly better recognized and understood.
Practical Implementation Options
Depending on the medium, different approaches may be considered:
- Images: EU icon or “AI-generated” label, placed in an unobstructed location (e.g., top right); supplemented by watermarks and metadata tags.
- Videos: Display the icon at the beginning, at regular intervals, and at least after every interruption (e.g., commercial breaks); supplemented by technical labeling in the metadata.
- Audio: An audible disclaimer at the beginning of the content; for long or live content, the disclaimer is repeated; supplemented by text-based labeling in the description and technical metadata.
- Text: Icon above or at the beginning of the text, or in the colophon.
Machine-readable labeling
In addition to the disclosure visible to humans, Article 50 (2) of the AI Regulation requires the provider of the generative system to label the output as artificially generated in a machine-readable format. This obligation does not directly apply to the company using the ; such a company can typically only ensure compliance through a contract with the tool provider or agency. For generative systems that were already on the market before August 2, 2026, the preliminary agreement on the “AI Omnibus” provides for a transition period until December 2, 2026.
The code generally provides for a multi-layered approach—digitally signed metadata (Sub-measure 1.1.1) and an invisible watermark (Sub-measure 1.1.2). However, a single layer of marking is sufficient if a generative system is embedded in a physical product and operates in a technically controlled, closed environment from which the output cannot escape—this may be relevant, for example, for embedded in-car or closed showroom systems. A single layer is also sufficient for free-form text, since free-form text cannot carry metadata.
The goal is to enable platforms, search engines, and other services to automatically detect AI-generated content. For businesses, this means that marketing agencies and content service providers should in the future often be required to
- provide the relevant metadata,
- not to remove the markings—the code prohibits the intentional removal or manipulation of these markings,
- comply with technical standards, and
- to document the origin of the content.
8. What exceptions are there?
The regulation provides for various exceptions. However, these are more narrowly defined than is often assumed.
Art and Satire
Art. 50 (4) of the AI Regulation only eases the disclosure requirement to the extent that the content is part of a “clearly artistic, creative, satirical, fictional, or analogous work or program”—and even then, a (non-intrusive) label remains required. The Code confirms that the obligation itself remains in place and that only the manner of disclosure is relaxed. Commercial advertising generally does not fall under this exemption.
Editorial Content
There are exemptions for certain journalistic contexts. However, these primarily apply to editorial activities in the public interest, where the text has been reviewed by a human and a natural or legal person assumes editorial responsibility. Traditional corporate communications, advertising, or product marketing are generally not covered by this provision.
9. How does this relate to competition law?
According to Article 50 (6) of the AI Regulation, the transparency requirements do not affect other provisions; in particular, the prohibitions on misleading practices under Sections 5 and 5a of the Unfair Competition Act (UWG) continue to apply. Accordingly, a visual representation is also considered misleading (Section 5 (4) of the UWG) if it misleads regarding essential characteristics of the product, such as design, accessories, or quality (Section 5 (2) of the UWG). The labeling requirement therefore does not replace an assessment under competition law. Companies should take particular care to ensure that AI-generated vehicle images do not suggest features, functions, or characteristics that are not actually present.
10. What should be done?
The time remaining until August 2026 should be used to establish appropriate governance structures.
(a) AI Inventory in Marketing
Companies should first identify which AI tools are being used, what content is being generated, which agencies are involved, and in which countries content is being published.
(b) Define Responsibilities
The marketing, compliance, data protection, and legal departments should work together to determine when disclosure requirements apply, which approval processes are in place, and who makes the final decision.
(c) Amending Contracts
Going forward, agency contracts should specifically include provisions regarding: disclosure of AI use, the origin of training data, labeling requirements, metadata standards, the non-removal of labels, and liability for violations.
(d) Develop a uniform labeling concept
International companies should define standards that are as uniform as possible and use the EU icon. Typical examples: “KI-generiert” (German), “AI-generated” (English), “Généré par IA” (French). A global standard reduces implementation effort and liability risks.
(e) Documentation and Record-Keeping
Regulatory authorities are expected to require companies to provide transparent documentation showing when AI was used, what content was affected, and how the labeling was implemented. Such documentation should become part of existing AI governance or compliance processes.
11. What’s Next?
The labeling requirements under Article 50 of the AI Regulation are among the first provisions of the AI Regulation that will have an immediate impact on broad business practices. They are likely to have a significant impact, particularly in the areas of marketing and communications. The risk of sanctions must be taken into account: Violations of the labeling requirement can be punished under Article 99 (4) of the AI Regulation with fines of up to 15 million EUR or 3% of global annual turnover.
Many detailed questions - particularly regarding the treatment of AI-generated product images and vehicle renderings - remain unresolved at this time. The draft Commission guidelines published on May 8, 2026, and the accompanying voluntary Code of Practice from June 2026 provide guidance but do not establish complete legal certainty and do not replace statutory obligations.
Until further clarification is provided by authorities and courts, a risk-based approach is therefore recommended: Companies should integrate transparency requirements into their AI governance at an early stage, define responsibilities, and, when in doubt, err on the side of labeling rather than omitting it. This not only reduces regulatory risks but also strengthens the trust of customers and business partners in the responsible use of AI.
