AI Act: What Does It Really Change for food Brands and Ad Producers?

Imagine this shot: a burger with sauce dripping down the side, cheese pulling at exactly the right moment, and a bun that looks like it came straight out of the oven. The problem is that no one actually cooked that burger. There was no kitchen, no food stylist, no photographer. There was a prompt. 

Not long ago, this kind of material raised questions primarily about aesthetics, ethics, and the authenticity of the message. From 2 August 2026, a new dimension has been added: obligations under Article 50(4) of the AI Act concerning the labelling of AI-generated or modified deepfake content. In practice, this means that some photorealistic materials created or modified using AI may require labelling. 

How broadly the new obligations will be interpreted remains a matter of debate, including among the relevant supervisory authorities for AI systems markets. The final Guidelines of the European Commission on the obligations of providers and deployers of AI systems regarding transparency, published on 20 July 2026, provide for a relatively broad understanding of the term "deepfake," indicating that it covers not only content resembling existing persons, objects, places, entities, or events, but also content that could plausibly exist, including not just images of people but also realistically depicted objects, places, and events. As a consequence of these guidelines, the labelling obligation will apply to a significantly larger share of commercial product content than many marketers currently assume. 

At Bites, we use AI tools every day, but we treat them as part of the production process, not a replacement for it. That is why we believe it is worth looking at the AI Act not solely through the lens of formal obligations, but as part of a broader system of regulation covering advertising, consumer protection, and intellectual property. 

This article explains what the AI Act actually changes, which obligations already existed, and what practical consequences the new rules may have for brands commissioning product photography and film. 

Three Layers of Law. The AI Act Is Just One of Them 

In discussions about the use of AI in advertising, the focus is almost always exclusively on the AI Act. That is understandable, but it is an incomplete picture. In practice, advertising materials depicting food are subject to several independent legal regimes simultaneously. 

The AI Act does not replace the other two layers. Even a correctly labelled AI-generated material may still violate food law if it depicts a product in a way that could mislead consumers about its actual characteristics, composition, or appearance. 

This is not a new principle. For years, Article 7 of Regulation (EU) No 1169/2011 has prohibited misleading consumers, including through the manner of presentation of a food product. This is complemented by rules on unfair commercial practices and the Act on combating unfair competition. 

Traditional food styling has long operated within the accepted boundaries of product idealisation. AI does not change those boundaries, but it significantly lowers the cost of creating images that may cross them. That is precisely why the new obligations under the AI Act must be analysed alongside the rules that have long governed how food is presented in advertising. 

Article 50 of the AI Act in Practice. When Can Food Content Require Labelling? 

The concept of "deepfake" raises the most questions today. In everyday usage, it is primarily associated with the faces of well-known individuals or fake video recordings. The definition in the AI Act is, however, broader. 

Under Article 3(60) of the AI Act, a deepfake is an AI-generated or manipulated image, audio recording, or video material that resembles existing persons, objects, places, entities, or events and may give the impression of being authentic. 

This final element is what matters most for the advertising industry. The rules do not focus on the creator's intent, but on how the content may be perceived from the perspective of an average recipient. 

It is worth noting that the scope of this definition is still being debated. The European Commission Guidelines point to a broad interpretation of Article 50, as noted above. The final boundaries of the obligations will depend on the definitive version of the guidelines and the practice of supervisory authorities. 

What Does This Mean for Product Content? 

Several typical situations can be identified in practice. 

A photorealistic product shot generated entirely by AI 

If a recipient could assume they are viewing a real photograph or recording of the product, the material may be subject to the transparency obligations under Article 50 of the AI Act. 

A real product placed in a generated environment 

This is one of the examples discussed in European debates about the AI Act. If an artificially generated setting gives the impression of a real place, such material may also require labelling. 

Modification of an existing photograph using AI 

Adding steam rising above a dish, a perfect drop of condensation on a bottle, or a spectacular "cheese pull" that did not actually occur may be considered a material intervention in the image. The assessment will depend on the nature of the changes and whether an average recipient could consider the result authentic. 

Graphics with a clearly illustrative character 

If a material clearly communicates that it is an illustration or artistic creation, the risk of it being classified as a deepfake is significantly lower. 

There is no simple test today that can answer each of these questions in advance. In practice, however, it is worth applying a precautionary approach: the more a material resembles a classic product photograph or a shot from a production set, the more significant the transparency obligations may become. 

Exception for Artistic Content 

The AI Act also provides for an exception covering materials of a clearly artistic, creative, satirical, or fictional character. 

This does not mean that every advertisement automatically benefits from this exclusion. The purpose of a typical pack shot or product photograph is to present the product as realistically as possible. The more a message relies on the authenticity of the product depicted, the more cautiously the artistic exception should be invoked. Moreover, the European Commission indicates in its Guidelines that the exception will generally not apply to advertisements, though this will depend on the specific circumstances. 

Only the practical application of the AI Act will show exactly where the boundary lies between advertising creative and material that should be labelled under Article 50. 

The Deepfake Definition Dispute. Why the Retail Industry Disagrees with the European Commission 

It is worth noting that the broad interpretation of Article 50 is not universally accepted today. 

On 18 June 2026, EuroCommerce, representing European retail and wholesale trade, addressed the European Commission with a call to clarify the draft AI Act guidelines. In the organisation's view, the current interpretation could subject a significant proportion of everyday marketing materials to labelling obligations, despite their non-disinformatory character. 

Among the examples cited were photographs of products displayed against computer-generated backgrounds. In practice, an analogous issue could arise for food content, for example a product depicted in a generated kitchen or restaurant setting. 

The most significant dispute concerns the criterion for assessing material. The AI Act and the European Commission Guidelines are based primarily on the objective perception of the message: what matters is whether the material could be considered authentic by its recipient. EuroCommerce, on the other hand, advocates greater weight being given to the creator's intent and the actual risk of misleading recipients. In the organisation's view, the obligations should not cover routine production activities that are not deceptive in nature. 

The European Commission Guidelines contain provisions and examples that at least partially address these concerns, indicating that the following are not deepfakes: 

• "films in which real actors (not modified by artificial intelligence) perform against a background generated by artificial intelligence (e.g. depicting a fictional stretch of prairie, historic buildings of ancient cities, etc.);" 

• "a real product (e.g. a car) shown in an advertisement against a background and in a setting generated by artificial intelligence, provided that the advertisement does not mislead consumers about the actual appearance, characteristics, and use of the product;" 

• "removal from recorded footage of bystanders visible in the background;" 

• "modifications or replacement of background for purely aesthetic purposes, compositions and arrangements of existing products, as well as changes of scale of images used in product advertisements or on packaging are likely to have only a minor impact on the recipient's perception of the authenticity and truthfulness of the advertisement and the product itself." 

Even if the scope of Article 50 is ultimately narrowed, this will not change the rules under food law. Material depicting a product in a way that could mislead consumers may give rise to legal consequences regardless of whether it is classified as a deepfake under the AI Act. 

Who Actually Has to Label? 

This is a question that in commercial practice is more important than the definition itself. The disclosure obligation under Article 50(4) of the AI Act covers content generated or manipulated by AI that may be classified as a deepfake, meaning realistic images, audio, or video that could mislead a recipient about what actually occurred in the real world. 

In advertising production, content is typically created by a producer or production studio, but it is the brand that publishes and uses it in communication. Reasonable and increasingly common market practice therefore assumes a division of roles between the parties: the producer informs the client about the use of AI in the form of an accurate, factual report, and the parties then jointly determine how the labelling obligation towards recipients will be fulfilled. This division of responsibilities should be set out explicitly in the contract, not left to assumption. 

For brands, this means a specific question to every production partner: will I receive documented information about what was generated in the material, which tool was used, and to what extent AI influenced the final image? Without that knowledge, it is difficult to properly fulfil one's own obligations. 

It is worth noting at this point the observation of attorney Jan Wiegner: 

Current State: The AI Act, Code of Practice, and Polish Legislation 

Several matters are worth keeping on the radar when planning advertising production in the coming months. 

Code of Practice on AI Content Labelling 

The European Commission is working on practical mechanisms to support the implementation of obligations relating to the labelling of AI-generated content. Although codes of practice of this type are voluntary, they may in practice become an important point of reference for regulators and the market. The direction is clear: the development of solutions such as content provenance metadata, digital labels, and mechanisms enabling the determination of whether material has been generated or modified by AI. 

European Commission Guidelines on Article 50 of the AI Act 

The practical interpretation of Article 50 will be one of the key topics for companies using generative AI. Particularly important will be the distinction between the mere use of an AI tool and a situation in which the recipient may be misled as to the authenticity of the depicted image, person, or event. 

Transitional Period and Obligations of Tool Providers 

The transitional arrangements being introduced relate primarily to providers of AI systems and technical content labelling mechanisms. This does not mean, however, that brands can disregard transparency towards recipients. Where content requires disclosure, the obligation to inform remains an element of responsible publication. 

Poland: Implementation of the AI Act 

The Polish legislator has established the national framework for the application of the AI Act, including the market supervisory system for artificial intelligence systems, through the Act of 3 July 2026 on artificial intelligence systems. The Act establishes a national body called the Commission for the Development and Security of Artificial Intelligence, which will decide on matters including financial penalties. The Act provides for the possibility of submitting requests to this body for individual opinions or explanations on matters within the scope of its activity. 

For the advertising industry, this means one thing: matters relating to the documentation of AI use are ceasing to be merely good practice and are becoming an element of risk management. 

Why Food Is a Special Case 

Food content has always been subject to stricter presentation rules than many other categories, because the image of food is in practice a communication about the product. 

A consumer looking at a photograph or video of a dish draws conclusions about its composition, size, freshness, quality, and preparation method. That is why rules governing the presentation of food have long treated the appearance of a product as an element of marketing communication, not merely aesthetics. 

AI amplifies this challenge. A generated image can create a product that cannot be purchased, an ingredient that is not in the recipe, or a culinary effect that is impossible to achieve in reality. 

The market is beginning to respond to this issue. Sales and delivery platforms are increasingly scrutinising rules regarding AI-generated food images, while advertising regulators are reiterating the basic principle: technology does not change the obligation of honest communication.  The control question remains simple: does the recipient, looking at the material, receive an accurate image of the product? 

The direction is clear. Brands planning food campaigns over the coming years should assume that transparency about AI use will be the norm, not the exception. 

Does AI-Generated Food Content Have Copyright? Work vs. Output 

The second matter that clients often discover only at the end of the production process is the legal status of the material itself. Polish copyright law protects works, meaning manifestations of human creative activity. A result generated solely by an algorithm, even on the basis of a human prompt, may not meet this criterion. 

In industry practice, such a result is sometimes referred to as a "product" or "output," meaning an effect of a tool's operation that may not be protected by copyright. This has a significant consequence: rights that have not come into existence cannot be transferred. 

This does not mean, however, that the material is entirely without protection. In certain situations, other legal bases may be relevant, for example provisions on combating unfair competition, particularly when someone appropriates the result of another party's work, investment, or organisational effort. That, however, is a different kind of protection from classic copyright. 

The good news is that an advertisement, spot, or campaign as a whole may still constitute a work, provided that its form results from the creative decisions of people: the director, director of photography, food stylist, set designer, editor, or creative team. The more documented, real human contribution there is to the process, the stronger the legal position of the material. 

Human control over the process is therefore not merely an element of production quality. It may determine whether the client receives a protected work or merely the output of a tool with uncertain legal status. 

AI Labelling Is Not Just Compliance. It Is a Matter of Trust. 

It is possible to meet all the formal obligations of the AI Act and still fail at the level of communication. Consumer research indicates that recipients evaluate content labelled as AI-generated differently, particularly when the label confirms their earlier suspicion that the material is artificial, exaggerated, or detached from reality. 

The conclusion is not: let us conceal the use of AI. 

Concealing the use of AI can be risky both legally and reputationally. The problem is not the technology itself, but the situation in which AI is used to create an image of a product that the consumer will not actually receive. If AI labelling becomes the standard, material must defend itself with something more important: quality, honesty, and creative control. 

The "AI" label on content that is truthful and well executed may be neutral information. The same label on an image of food that promises a product which does not exist in reality becomes a warning signal. 

AI Act in Food Marketing: Compliance Checklist Before Publication 

Six questions to ask yourself and your production partners: 

Asset audit. Which assets in your library and current campaigns contain elements generated or modified by AI? Where will they be published? 

Authenticity test. Could a recipient consider the material a real photograph or recording of the product? If so, check whether labelling is required. 

Division of obligations in contracts. Do the production contracts clearly specify who will disclose the use of AI and how? 

AI report from the producer. Do you receive information about what was generated, using which tool, and to what extent? 

Metadata and process documentation. Does the production process preserve information about AI use and document the human creative contribution? 

Food layer. Regardless of the AI question, does the image suggest product characteristics that the product does not actually possess? 

In Conclusion 

At Bites, we repeat that the value of professional production does not lie solely in the creation of the image. It lies in control: deliberate decisions, brand consistency, and accountability for the result. 

The AI Act reinforces exactly that logic. It will reward those who know what ended up in the frame, why it ended up there, and who can document that process. AI can accelerate production, but it does not replace accountability for the message. 

In food content, what matters most remains the same: the image must be not only attractive. It must be true.