Deadline August 2nd: New AI law comes into force – Why almost every company now has to label its AI texts

AI Regulation warning trap: Most companies are ignoring this new obligation starting in August

The 4 mandatory requirements you must implement from August: Millions in fines are threatened – What will change for users of ChatGPT & Co. on August 2nd

On August 2, 2026, things will get serious for the European economy: Article 50 of the new AI Regulation (AI Act) will come into full legal effect. What is still mistakenly dismissed in many boardrooms and marketing departments as a purely technological issue, or even something postponed to the distant future, turns out, upon closer inspection, to be a ticking time bomb for everyday practice. Anyone who uses generative artificial intelligence – be it in the form of website chatbots, automated marketing texts, or AI-generated images – without regulation and proper labeling risks hefty fines of up to €15 million from this date onward.

The era of reckless experimentation is definitively over; now begins the phase of strict compliance. The following article examines in detail why the EU's new transparency obligation is not merely a paper tiger and which four concrete obligations now apply. Learn exactly when you are required to label your AI content, what the difference is between providers and users, and why the best protection against legal warnings lies not in technical filters, but in sound internal processes.

When the machine must fall silent: Why August 2, 2026 will be the moment of truth for European companies

A law that nobody takes seriously until it strikes

Article 50 of the European Regulation on Artificial Intelligence will fully enter into force on August 2, 2026. For most CEOs, marketing directors, and communications managers in Germany, Austria, and Switzerland, this date initially sounds like just another bureaucratic footnote in an already overloaded regulatory calendar. However, a closer look quickly reveals that this is one of the most consequential provisions of the entire AI Regulation, because it affects not only high-risk applications but virtually every company that uses any form of generative artificial intelligence in its communication, marketing, or customer service. The public debate surrounding the so-called Digital Omnibus, which postponed large parts of the high-risk obligations to 2027 and 2028, has led many decision-makers to mistakenly believe that the entire AI Act has been weakened or postponed. This is simply not the case. The transparency and labeling requirements of Article 50 are explicitly excluded from this postponement and will come into effect on the originally scheduled date.

Anyone operating a chatbot on their website, publishing an AI-generated press release, using a voice assistant in customer service, or creating marketing materials with generative image software will, from this date onward, be operating in an area where ignorance is no longer an excuse. The fines for violations can reach up to €15 million or 3 percent of global annual revenue, whichever is higher. For a medium-sized company with a group revenue of €200 million, this effectively means the maximum fine of €15 million, as the percentage is lower in this case. For a so-called hidden champion with €1.5 billion in revenue, however, the maximum penalty can rise to €45 million, because above a certain revenue threshold, the percentage exceeds the fixed upper limit. These figures make it clear that this is by no means a symbolic sanction, but rather a Damocles' sword hanging over all corporate communication practices.

Four duties, one principle: transparency instead of deception

Anyone who delves into Article 50 will quickly encounter an important distinction that is often overlooked in superficial debates. The provision consolidates four independent obligations that affect different actors and must be considered separately in practice. The first obligation requires providers of AI systems designed for direct interaction with humans to clearly and unambiguously inform users from the very first contact that they are interacting with a machine and not a human. A chatbot that introduces itself with a human first name and provides no explicit indication of its artificial nature does not meet this requirement, even if a corresponding clause is hidden in the fine print of its terms and conditions.

The second obligation applies to providers of generative AI systems that generate synthetic audio, image, video, or text content. They must ensure that their output is marked in a machine-readable format so that technical systems can recognize that it is artificially generated or manipulated content. This so-called watermarking requirement primarily affects the large technology providers themselves and was postponed by four months to December 2, 2026, as part of the Digital Omnibus. This postponement is often mistakenly perceived by the public as a general postponement of all obligations.

The third obligation concerns the use of emotion recognition or biometric categorization systems. Anyone using such technologies, for example in the analysis of customer conversations in call centers or for biometric access control, must inform the individuals concerned about this function and, if necessary, obtain their consent. This transparency obligation applies in addition to the existing data protection requirements of the General Data Protection Regulation (GDPR).

The fourth obligation, and arguably the most significant in business practice, concerns the labeling of so-called deepfakes and AI-generated or AI-manipulated texts intended to inform the public about matters of public interest. This is precisely where the real complexity begins for brands, editorial teams, and communications departments, because numerous vague legal terms converge here, the practical interpretation of which is only clarified by guidelines from the European Commission and a voluntary code of conduct.

The term "deepfake" is broader than many people realize

A key misunderstanding in the public debate concerns the term "deepfake" itself. Colloquially, many people think of manipulated videos of politicians or fake celebrity photos. However, the legal definition, as it emerges from the draft Commission guidelines, is considerably broader. According to the regulation, a deepfake exists if image, audio, or video content has been artificially created or manipulated, if it resembles real people, objects, places, facilities, or events, if it could be mistakenly perceived by a person as genuine or truthful, and if this effect is objectively assessed from the perspective of the audience. An intent to deceive on the part of the creator is explicitly not a prerequisite for classification as a deepfake.

This broad interpretation has significant practical consequences. It covers not only obvious manipulations of well-known personalities, but also realistic-looking AI-generated product images, fabricated press photos, or synthetic stock images that suggest authenticity even though they were never photographed. A company that uses a deceptively realistic AI-generated image of a family at the breakfast table for a campaign is thus potentially subject to the labeling requirement, while an obviously stylized cartoon image or a recognizably artistic illustration typically remains exempt. Furthermore, purely technical audio processing such as noise reduction or volume normalization is not covered, as this does not constitute content manipulation within the meaning of the regulation.

When a text becomes a matter of public interest

The question of which texts qualify as serving to inform the public about matters of public interest is similarly complex. According to the guidelines, three conditions must be met cumulatively. The text must be published, meaning it must be disseminated to an indefinite, large number of people, whether through an article, a push notification, a news ticker, or a public social media post. It must also serve to inform the public, meaning it must convey knowledge, opinions, or facts, and not merely provide entertainment. Finally, it must address a matter of public interest, which includes topics from politics, administration, health, the environment, economics, science, education, security, and cultural issues of general relevance.

For business practice, this means that classic editorial content such as election coverage, legislative analyses, economic and health news, or official warnings are regularly included. Purely marketing texts, internal memos, and entertainment formats, on the other hand, generally do not fall under this category, but may, in individual cases, become subject to it if they address topics of general relevance. A particularly sensitive example is press releases on sustainability issues, crisis communication, or product recalls, as these topics are explicitly identified by the Commission's guidelines as areas of public interest, even if they originally stem from traditional corporate communications.

The editorial exception as a lifeline with high hurdles

For texts, but not for deepfakes, the regulation provides a significant exception to the labeling requirement. This requirement is waived if human review or editorial control takes place before publication and if editorial responsibility is assumed by a clearly identifiable natural or legal person. At first glance, this exception appears to be a simple way out of the labeling requirement, but the requirements for genuine editorial review are considerably stricter than many companies initially assume. The draft Commission guidelines make it unequivocally clear that a mere spelling or formatting check is insufficient. What is required is a deliberate content review for accuracy, plausibility, and source, combined with a genuine opportunity to amend or reject the text entirely.

This high threshold has practical consequences that extend far beyond mere text production. An article draft generated by artificial intelligence that is actually reviewed by an editorial team falls under the exception. So does a routine AI-generated report on weather conditions, financial statements, or sports results that undergoes at least a plausibility check. However, the exception does not cover comments written almost entirely by AI that have merely been linguistically smoothed, automated translations without any content review, or automated aggregation tickers without any prior review. For companies with extensive content production, this means that the question of labeling requirements ultimately transforms into a question of robust internal processes: Who checks what, on what basis, and how can this review be proven to a regulatory authority or a court in the event of a dispute?

Providers or users: The underappreciated dual role of small and medium-sized enterprises (SMEs)

The regulation fundamentally distinguishes between providers who develop an AI system or market it under their own name, and users who employ such a system on their own responsibility within the scope of their professional activities. In theory, this distinction seems clear, but in business practice, the boundaries are becoming increasingly blurred. As soon as a company significantly modifies an existing AI model, republishes it under its own name, or integrates it into its own chatbot, voice assistant, or digital avatar, it can inadvertently slip into the role of a provider, with all the associated additional obligations. A company that merely uses the standard interface of a well-known AI provider clearly remains a user. However, a company that integrates a model interface behind its own workflows and input prompts and presents the product under its own brand name already assumes provider obligations regarding interaction instructions. Finally, anyone who trains a model on their own data and offers it internally and externally as an independent product bears full provider responsibility, including the technical labeling requirements.

This dual role is increasingly becoming a core problem in consulting practice because many medium-sized companies have unknowingly slipped into the role of provider, for example through so-called white-label chatbots, self-developed voice agents, or branded digital avatars in marketing campaigns, without being aware of the resulting additional obligations. Those who fail to clarify this role issue early on risk facing a double argumentation problem in the event of an official audit, because neither the provider nor the user obligations have been fully met.

Xpert.Digital is a data-driven B2B industry hub led by Konrad Wolfenstein . The company acts as an external, quasi-in-house solution for industrial partners, closing operational gaps in marketing, content, and sales – without requiring additional resources on the client side.

More information here:

• The quasi-in-house solution: How Xpert.Digital closes operational gaps in B2B marketing and sales – Smart Content-Driven Business

Detailed analysis: When exactly must an AI-generated image or AI-supported content be labelled as such?

The practically crucial question for marketing and communications departments is under what specific circumstances an individual image, video, audio clip, or text actually requires visible labeling. Based on the Commission's previous clarifications and the relevant literature, a practical framework for assessment can be derived, which can be broken down into several successive questions.

The first question concerns the origin of the content: Was the material generated entirely or substantially by artificial intelligence, or is it merely a case of supportive use, where a human takes an AI-generated draft, substantially revises it, and thereby significantly shapes it? In the latter case, the labeling requirement is generally waived because the artificial intelligence merely served as an aid, and the impression is not created that a contribution was written exclusively by a machine.

The second question concerns publication: Is the content shared with an unspecified, wider public, or does it remain exclusively for internal or private use? Purely internal applications, such as a company-internal knowledge assistant or an AI-supported design that never leaves the company, generally do not trigger any labeling requirements. However, as soon as an employee incorporates an internally generated AI text into an external press release, a sustainability report, or a public response to a product recall, the obligation can immediately apply.

The third question concerns the content category in the case of images, audio, or video material: Does the content resemble a real person, object, place, or event in a way that could objectively create an impression of authenticity for the audience? A purely illustrative, clearly stylized standard image without any reference to real people or events typically does not fall under the definition of a deepfake. However, a realistic-looking lifestyle image with AI-generated people that could resemble real people, or a fabricated press photo of an actual event, typically triggers the labeling requirement.

The fourth question concerns the thematic classification of texts: Does the text inform the public about a matter of public interest, such as health, the environment, safety, politics, or economic issues of general relevance, or is it pure product advertising without any informative character? Purely sales-oriented texts generally remain outside the scope of application, while native advertising with an editorial feel on health or financial topics can certainly fall within the scope.

The fifth and final question concerns editorial control: Did a genuine content review take place before publication, coupled with a clearly documented assumption of editorial responsibility by an identifiable person or organization? Only if this review actually took place and is demonstrably documented does the labeling requirement for texts in the public interest cease. This exception explicitly does not apply to deepfakes, meaning that corresponding image, audio, or video content must generally be labeled even after editorial review, provided it meets the aforementioned criteria.

In addition to this content review framework, the question arises of concrete implementation. The labeling must be clear and easily distinguishable, visible or audible at the latest upon first encountering the content, understandable even to laypersons and particularly vulnerable groups, accessible to all, and not hidden in terms and conditions or submenus. For images, labeling directly in the caption or as an overlay on the image itself is recommended so that the notice remains visible even when shared on social networks. For audio content, a spoken notice at the beginning of the recording is recommended, supplemented by a visual label for embedded players. For videos, continuous labeling, or at least repeated labeling at the beginning, is considered appropriate for live broadcasts, while for short videos, continuous labeling from the beginning is advisable, and for longer formats, labeling at the beginning, at regular intervals, and possibly in the end credits is recommended. For texts, a uniformly chosen, clearly recognizable position is recommended, such as above the actual text, in the heading or in a corresponding note block at the beginning of the article, whereby for texts that are only partially AI-generated, marking the relevant part of the text is sufficient.

A special exemption applies to works that are clearly artistic, creative, satirical, or fictional. In these cases, the transparency requirements are limited to appropriate disclosure that does not impair the enjoyment of the work. For example, a note in the opening or closing credits of a feature film is typically sufficient, and continuous displays throughout the entire runtime are not required. It is important to note, however, that this exemption applies solely to the labeling requirement itself and does not remedy any other legal infringements. Personal rights, copyrights, trademark rights, and the provisions of the General Data Protection Regulation (GDPR) remain completely unaffected and must be respected regardless of the labeling issue.

Why enforcement is more like the General Data Protection Regulation than a technical identification system

An often overlooked aspect of the new regulation concerns how it can be enforced when it becomes increasingly difficult to clearly distinguish AI-generated from human-created content. The answer lies less in seamless automated detection than in a reversal of the burden of proof and greater reliance on market mechanisms. In cases of doubt, companies must be able to demonstrate that they acted in accordance with the law to the best of their knowledge and belief, which significantly increases the importance of documentation, metadata, and traceable internal processes. Whether a single piece of content can be technically and unequivocally identified as AI-generated is of secondary importance. What matters is whether a company can provide comprehensive evidence in a dispute regarding which tool, version, and internal approval process were used to create a specific piece of content.

This mechanism is strongly reminiscent of the enforcement practice of the General Data Protection Regulation (GDPR), where the concrete interpretation of many vague terms only developed through individual court cases, competitor lawsuits, and official warnings. It is expected that the rules of Article 50 will be defined in a similar way through concrete precedents, with violations not only being prosecuted by national market surveillance authorities but also being actively initiated by competitors under competition law. In Germany, the Federal Network Agency (Bundesnetzagentur) is to become the central market surveillance authority. It has been operating its own advisory service for companies since summer 2025 and has been systematically building its supervisory structure since the beginning of 2026. While a flood of penalty notices is not expected immediately after the provision comes into force, random checks are anticipated in the medium term in particularly exposed sectors such as consumer services, banking, insurance, public administration, and healthcare. Companies that cannot demonstrate robust internal processes and complete documentation will be significantly more vulnerable in this environment than those that implemented structured approval procedures early on.

Governance instead of panic: The real core of the new duty

Anyone who considers the various facets of Article 50 together quickly realizes that the real challenge lies not in simply affixing a label, but in establishing robust internal governance structures. In the future, companies will have to be able to provide complete documentation of how specific content was created, which technical systems were used, and whether the necessary review and approval processes actually took place. This question thus shifts from a purely legal footnote to a central task for marketing and communications departments, which must work closely with the legal department, as well as with information technology and senior management. Key components of this governance include a complete inventory of all AI systems actually used in the company, including so-called shadow AI (i.e., tools used without authorization by individual employees), a clear allocation of which obligation under Article 50 applies to which use case, a technical implementation of visible and audible labels in the appropriate places in the content management system, adapted contracts with freelancers and agencies that define clear responsibilities for editorial review, and regular training for all employees working with artificial intelligence.

Current market data underscores the urgency of this task. The proportion of German companies with 20 or more employees actively using artificial intelligence has more than doubled within a year, rising from 17 to 41 percent, while for companies with 500 or more employees, the figure is already over 60 percent. The most frequent applications in SMEs are precisely those areas where Article 50 directly applies: text creation, document analysis, and customer communication via chatbots. These figures demonstrate that this is by no means a fringe topic for a few technology companies, but rather a cross-industry reality affecting the vast majority of the German economy.

Labeling requirements as a catalyst for more professional content processes

The new transparency requirement can ultimately be understood as a kind of stress test for the maturity of corporate content processes. Companies that already have clear responsibilities, documented approval processes, and a central directory of their AI tools will manage the transition on August 2, 2026, relatively smoothly. In contrast, companies that have integrated artificial intelligence into their communication and marketing processes in an uncontrolled manner and without clear governance face a significant catch-up task that can hardly be accomplished in the remaining weeks without a structured approach. It is crucial to recognize that technical labeling alone is insufficient if there are no traceable processes in place that can be presented in the event of a dispute. Ultimately, the regulation forces companies to do something that, from the perspective of professional brand management, was long overdue: to consciously manage and document the use of artificial intelligence in their own communication and thereby also to make it more responsible internally. Those who seize this opportunity will be able to turn the new obligation not only into a compliance risk, but into a real competitive advantage in the form of increased trust among customers, partners and the public.

☑️ Our business language is English or German

☑️ NEW: Correspondence in your native language!

 

I and my team are happy to be available to you as your personal advisor.

You can contact me by filling out the contact form here wolfenstein@xpert.digital:or simply call me at +49 7348 4088 965. My email address is

I'm looking forward to our joint project.

 

 

☑️ SME support in strategy, consulting, planning and implementation

☑️ Creation or realignment of the digital strategy and digitization

☑️ Expansion and optimization of international sales processes

☑️ Global & Digital B2B trading platforms

☑️ Pioneer Business Development / Marketing / PR / Trade Fairs

In industrial B2B, sustainable business relationships rarely emerge overnight. They develop step by step – through visibility, professional relevance, recurring touchpoints, and growing trust. Xpert.Digital's 4-stage model addresses precisely this: It offers a structured path that begins with a manageable entry point and can evolve into deeper collaboration in business development if needed.

Instead of relying on loud marketing promises, this model puts the relationship at the forefront. Companies start with clearly defined, easily calculable measures and then decide, based on their own experience, how far they want to expand the collaboration. A key factor for this undisturbed trust-building process: The platform completely avoids annoying advertising ads, so the editorial focus remains solely on the companies' expertise.

More information here:

• From visibility to trust: Your scalable path with Xpert.Digital