Towards an Era Where "AI-Made" Cannot Be Hidden: What New EU Rules Impose on Japanese Companies and Influencers

On August 2, 2026, a new rule with the potential to change how information appears on the internet began to take effect in the European Union. Under the EU's comprehensive artificial intelligence regulation, known as the "AI Act," transparency obligations have been applied to content created by generative AI and AI systems that respond on behalf of humans.

The targets are not just so-called deepfakes. A wide range of content and services that generative AI creates, such as images, videos, audio, text, corporate chatbots, AI avatars, and voice agents, which humans might perceive as genuine, are involved.

From the headlines of news articles, it might be understood as "In the EU, everything created by AI must have a conspicuous label." However, the actual system is a bit more complex. The rules are broadly divided into "those providing AI systems" and "those using AI systems in business."

It's Not a System Where the Same Label Is Applied to All Creations

First, companies providing AI systems like ChatGPT, Gemini, and image generation services are required to make it possible for machines to determine that the creations were made or processed by AI.

Specifically, the idea is to establish a mechanism that combines metadata, electronic signatures, and detectable watermarks, allowing the artificial origin to be confirmed even if images or videos are reposted on other sites. It does not merely mean displaying a large warning text to the human eye at all times, but rather leaving machine-readable marks on files and content is crucial.

Additionally, chatbots and AI agents that converse directly with users must be designed so that it is clear they are not human. Even if natural language is returned at an inquiry window, it is necessary to avoid a situation where users mistakenly believe they are conversing with a human representative.

On the other hand, companies and businesses that produce and publish advertisements, articles, videos, etc., using AI are required to display specific content in a way that humans can recognize.

Particularly important are deepfakes that make real people, places, objects, organizations, events, etc., appear genuine. If AI-created or processed videos or audio could be mistaken for those actually filmed or recorded, clear labeling is required at the first point of contact with the viewer.

Texts that could influence social debates in areas such as politics, administration, justice, public safety, public health, environment, economy, finance, science, and culture may also be subject to this. If AI-generated or processed texts are published without sufficient human verification or editing, this must be clearly indicated.

Conversely, if a system is in place where humans with expertise verify the content, check sources, and can amend or halt publication, even highly public texts may be exempt from the display obligation. However, merely correcting typos or grammar does not constitute substantial human review.

In other words, the question is not just "Was AI used even once?" but rather how much of the content was created by AI, how much responsibility humans took in verifying it, and whether there is a possibility that viewers might mistake it for genuine.

Influencers Are Not Necessarily Exempt Just Because They Are Individuals

Under EU rules, individuals using AI for personal and non-professional purposes are generally exempt from the obligations on the business side.

For example, posting images generated as a hobby on a personal account is usually not considered a business activity. However, if economic benefits are obtained through advertising revenue, corporate projects, product sales, membership services, or continuous donations, the activity may be deemed professional or commercial.

It is not determined solely by the number of followers. Influencers, video distributors, affiliates, and freelance creators who earn revenue continuously may be treated as "users" of AI in their business.

If an AI model resembling a real person is used in an advertisement or if AI voices are used to narrate product descriptions not spoken by the person, clear labeling is likely required.

On the other hand, fantasy images that are clearly not real or works that viewers understand as fiction do not always require prominent warning labels. For art, creation, satire, and fiction, display methods that do not hinder the appreciation of the work are permitted.

This point is important for Japanese advertising and content production companies. When creating campaigns for the EU, it is necessary to organize the reality, potential for misunderstanding, purpose of use, and target audience, rather than making a uniform judgment based solely on the fact that "AI images were used."

Why Did the EU Mandate Labeling?

The background is the reality that generative AI can now create "realistic fakes" at low cost and in large quantities.

Previously, synthetic videos and audio required advanced editing skills and equipment. However, now even general users can quickly create speeches not actually made by politicians, fictitious product endorsements by celebrities, explanations by non-existent experts, and fake disaster footage.

In corporate advertising, it has become possible to feature non-existent doctors, researchers, consumers, and influencers to emphasize the effectiveness and popularity of products. If viewers are unaware that the content is AI-generated, they may receive it as real testimonials or expert recommendations.

What the EU aims to protect is not just copyrights or personal information. It encompasses consumer decision-making when choosing products, decision-making regarding elections and policies, trust in news, and human-to-human communication itself.

The basic idea of the transparency rule is not to "ban fakes" but to "reduce the act of disguising artificially created items as genuine and distributing them."

Mixed Reactions on Social Media

The new rules have sparked active discussions in European tech communities and on platforms like LinkedIn and Reddit. Although posts on social media are not opinion polls and do not represent society as a whole, they provide clues to the questions and concerns users and practitioners have.

Notable are the welcoming voices saying, "At least we should be informed whether it's AI or human."

Particularly regarding customer service and phone support, there is strong dissatisfaction with not knowing whether the counterpart is AI or human. Opinions such as "Not only images and videos but also the fact that the conversation partner is AI should be displayed" are seen, and there are reactions that appreciate the fact that chatbots and voice agents are also included in the current rules.

Due to the increase in political fake videos and fake advertisements by celebrities, some posts suggest that "similar systems are needed in other regions like the United States." Many users positively perceive the EU's response as an attempt to catch up with legal changes in technology.

On the other hand, there are voices expressing concern about the ambiguity of the display method.

Questions such as "What size should the label be to be sufficient?" "Is it enough to display it small in the corner of the screen?" "Is it necessary to make it understandable to those who start watching from the middle of the video?" arise. Although the EU provides icons for labeling, their use is not mandatory, and businesses can choose other methods.

While there is flexibility, there are indications that it is difficult to understand to what extent implementation will be recognized as "clear labeling." For small production companies and individual creators, the burden of understanding all aspects of law, design, and technology is not small.

Another prominent concern is whether it will become a "repeat of cookie banners."

If AI labels are attached everywhere like cookie consent screens displayed every time a website is opened, users might reflexively ignore them without checking the content. If there are too many warnings, the difference in importance between truly dangerous deepfakes and merely generated background images may become indistinguishable.

Even if it only says "AI used," it is unclear whether the person's face was completely synthesized or if only the brightness of the photo was adjusted. The existence of a label and users correctly understanding the content are not the same.

There are also persistent doubts about the effectiveness of enforcement. The issue is how far content created outside the EU and disseminated through anonymous accounts can be tracked. Metadata may be lost due to image compression, screenshots, reposting, and malicious users may intentionally delete it.

On social media, reactions such as "How to deal with a large amount of content created abroad" and "Only companies that comply with the display bear the burden, while malicious posters ignore it" are also emerging.

Will People Stop Being Deceived If There Is a Label?

AI labeling is important, but it alone does not solve the problem of misinformation.

Previous research has shown that users who see a label indicating AI generation tend to evaluate the information lower than before, but cases have also been reported where persuasion and sharing behavior do not necessarily decrease significantly.

Moreover, there is a danger that users might overconfidently believe that "items not labeled as AI-generated are genuine." Acts of attaching false explanations to genuine images taken by humans or human-created misinformation cannot be prevented by AI labels alone.

The possibility of incorrect labeling is also problematic. If an actual photo is judged as AI-generated, the credibility of the photographer or news agency may be damaged. Conversely, if cleverly generated items cannot be detected, trust in the labeling system itself may decline.

Therefore, AI labels should be considered "additional information for judging the production process" rather than "marks guaranteeing authenticity." Ultimately, reliability needs to be judged comprehensively, considering the information source, editorial responsibility, verification method, and the existence of the sender.

Japan Chooses a Different Path from the EU

In Japan, the first AI-specific law, "Law for the Promotion of Research and Development and Utilization of Artificial Intelligence Technology," was enacted in 2025 and fully implemented in September of the same year.

Japan's AI law also emphasizes the need to ensure transparency for proper research, development, and use. However, when comparing the current system structure, it differs in nature from the EU's regulation, which cross-sectionally defines the display methods and targets of generated content and imposes hefty penalties for violations.

Japan has a strong stance of responding to risks by promoting AI research, development, and utilization while combining government guidelines, information gathering, investigations, guidance and advice to businesses, and existing laws.

If the EU is a "regulatory type" that imposes specific obligations on businesses, Japan is closer to a "promotion and cooperation type" that emphasizes cooperation between the public and private sectors and guidelines.

This difference is not a simple matter of which is correct. Strict regulations can lead to consumer protection and trust assurance, but they may increase response costs and suppress the activities of small and emerging companies. A flexible system makes it easier to advance technological development, but it tends to leave the response to malicious deepfakes and fake advertisements to businesses.

What is important for Japan is not to judge based solely on the domestic system. When Japanese companies sell products to EU consumers, operate EU-targeted websites and advertisements, and provide AI services through local subsidiaries or agents, they cannot remain unrelated to EU rules.

Five Necessary Responses for Japanese Companies

First, it is necessary to understand where generative AI is being used within the company.

Generative AI is often introduced individually in multiple departments, such as advertising images, product descriptions, SNS posts, recruitment videos, inquiry responses, translations, and narrations. If there is "on-site use" that the company is not aware of, it cannot make judgments about display obligations.

Second, it is necessary to confirm what marks AI providers are attaching to their creations.

Even if metadata is attached at the time of generation, it may be deleted by image editing software, advertising distribution systems, CMS, or uploads to SNS. It is necessary to verify whether AI-derived information is retained in the process from production to publication.

Third, it is necessary to define the meaning of human verification within the company.

It may not be sufficient to treat it as "human-verified" just because a person has read the text once. It is important to clarify fact-checking, source verification, legal and advertising review, correction authority, and publication responsibility, and to record who bears the final responsibility.

Fourth, it is necessary to review contracts with advertising agencies, production companies, and freelancers.

Clauses are required to confirm whether AI is used in deliverables, which tools were used, whether real people or places are mimicked, and whether necessary labels and metadata remain. Just because the contractor produced it does not necessarily absolve the advertiser or ordering company of responsibility.

Fifth, it is necessary to design displays not just as a legal burden but as a means of demonstrating trust.

Rather than just displaying "AI used," specifying the scope of use, such as "background image generated by AI," "voice synthesized," or "AI used to create text drafts, fact-checked by the editorial team," may lead to consumer understanding.

The purpose of the display is not to apologize for using AI. It is to explain what was artificially created and where humans take responsibility.

What Is Required of the Media Is "Human-Responsible Editing"

For news sites and information media, the new rules hold particularly heavy significance.

A system that uses generative AI to create a large number of articles and only checks headlines and typos by humans may be judged as not having sufficient editorial management for highly public texts.

The important thing is not to avoid using AI itself. It is whether there are humans who confirm information sources, compare different perspectives, correct errors, and decide whether to publish.

Even if AI assists with research, translation, composition, and summarization, if an editor responsible for the final content is functioning, transparency and efficiency can be balanced. Conversely, if only human names are formally placed and AI is effectively auto-publishing, it is difficult to gain reader trust.

For Japanese media, showing "who is responsible for the information" rather than "whether AI was used" will become a competitive edge in the future.

The Competition for Global Standards Begins

The EU's regulation may not be confined to within its borders.

For global platforms and AI companies, it may be more efficient to adopt the standards of the strictest market as a global common specification rather than operating different systems for each region. Just as the EU's personal data protection regulations have influenced privacy responses of companies in various countries in the past, the display of AI-generated items may become a de facto international standard.

In China and South Korea, systems for displaying AI-generated content and AI advertisements are also advancing, making it a global policy issue to indicate the origin of AI.

What Japan should consider in the future is not just whether to introduce the same laws as the EU. It is necessary to