What Is the EU Code of Practice on Transparency of AI-Generated Content? Corporate Responses to Article 50 of the AI Act

On July 8, 2026, the European Commission, and on July 9, the AI Board, published an opinion assessing the "Code of Practice on Transparency of AI-Generated Content" as an adequate means of meeting the obligations under Article 50 of the AI Act. The code is a voluntary code of conduct whose final version was published on June 10, 2026, after multi-stakeholder consultations facilitated by the AI Office. On June 11, OpenAI announced a statement supporting the code. Because the obligation to label AI-generated content under Article 50 of the AI Act begins to apply on August 2, 2026, the publication of the code's assessment and OpenAI's statement of support are part of a series of developments immediately before that date of application. This article organizes the legal nature of the code and the Commission Opinion, and then explains the obligations that Article 50 imposes on providers and deployers, the difference between participating in the code and complying with the law, and the items Japanese companies should check.

What Was Published

The official name of the code is the "Code of Practice on Transparency of AI-Generated Content." Drafting began in September 2025, led by six independent experts appointed by the AI Office, and the final version was published on June 10, 2026, after three rounds of consultation involving more than 187 stakeholders, including industry, academia, civil society, rightsholder organizations and member state representatives. The code consists of two parts: a chapter on marking and detection for providers of generative AI systems, and a chapter on labeling for deployers.

The assessment by the European Commission and the AI Board is not the publication of the code itself but an assessment of the code's adequacy. The assessment document states that the code adequately covers the obligations under Article 50(2), (4) and (5) of the AI Act and facilitates their effective implementation, while also expressly stating that "Adherence to the code does not constitute conclusive evidence of compliance." I think participation in the code does not by itself legally establish compliance with the law.

The code is based on the framework of Article 50(7) of the AI Act, which provides that the AI Office is to encourage and facilitate the drawing up of codes of practice to facilitate the effective implementation of the obligations regarding the detection and labeling of AI-generated or manipulated content, and it is a separate regime from the code of practice for providers of general-purpose AI models provided for in Article 56. Because this code remains positioned as voluntary practical guidance supporting the implementation of the transparency obligations under Article 50, I think the two should not be treated as regimes with the same legal effect.

The Obligations Under Article 50 of the AI Act

The date of application of Article 50 is August 2, 2026, under Article 113 of the AI Act. The article imposes different obligations on several types of entities, and the practical starting point is not to confuse the distinctions drawn in its text.

Providers have the following two types of obligations. Paragraph 1 requires providers to design and develop AI systems intended to interact directly with natural persons so that users are aware that they are interacting with AI. Exceptions apply where this is obvious from the circumstances and for uses authorized by law to detect, prevent or investigate criminal offenses and the like. Paragraph 2 requires providers of AI systems that generate synthetic audio, image, video or text content to ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. They are also required, to the extent technically feasible, to ensure the interoperability and robustness of the marking.

Deployers have the following two types of obligations. Paragraph 3 requires deployers using an emotion recognition system or a biometric categorization system to inform natural persons of the operation of the system. Paragraph 4 requires deployers who generate or manipulate deep fakes (artificially generated or manipulated image, audio or video content) that appreciably resemble existing persons, objects, places, entities or events and could falsely appear to be authentic to disclose that fact. A similar disclosure obligation applies to AI-generated text published for the purpose of informing the public on matters of public interest, with an exception where it has undergone human review or is subject to editorial responsibility. Paragraph 5 is a provision requiring this information to be provided in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.

The code covers paragraphs 2 and 4, as well as paragraph 5, which sets the timing and manner of provision. Paragraph 1 (notification of interaction) and paragraph 3 (notification of emotion recognition and the like) are outside the scope of the code, and this also shows that the subject of this assessment is not Article 50 as a whole but is limited to the parts concerning the marking and labeling of AI-generated content.

Participating in the Voluntary Code and Complying with the Law Are Different Things

The most important point for companies to check is that signing the code is not a substitute for obligations under the AI Act. Each obligation under Article 50 is a legal obligation, and regardless of whether they sign the code, the providers and deployers concerned must fulfill these obligations from August 2, 2026. The code is voluntary guidance showing in practical terms how to fulfill the obligations, and not signing it is said not to mean non-compliance in itself. It has also been explained, however, that businesses that have not signed may be more likely to receive requests for information from supervisory authorities.

Businesses wishing to sign the code must submit to the European Commission, by 18:00 (Central European Summer Time) on July 22, 2026, a form signed by a senior executive with sufficient authority in order to be included in the initial list of signatories. The initial list of signatories is scheduled to be published before August 2, the date of application of Article 50. Signatories are said to be able to use the code as a means of demonstrating compliance in light of this adequacy assessment by the European Commission and the AI Board, but I think this is a practical means of demonstration premised on substantively meeting the requirements of Article 50, and it does not legally create a formal presumption of conformity.

Points of Contact and Responses for Japanese Companies

The AI Act, including Article 50, may apply even to businesses without an establishment in the EU. Article 2 of the AI Act provides that even providers and deployers located in a third country are covered where the output of the AI system is used in the EU. Accordingly, where a Japanese company provides chatbots or image or audio generation services to consumers in the EU, or where its subsidiaries or business partners in the EU use such AI systems for the EU market, and also where the output of services provided from Japan is used in the EU, it is necessary to check case by case whether the obligations of a provider or deployer apply.

On the other hand, current Japanese law does not appear to contain any provision generally imposing a labeling obligation for AI-generated content similar to Article 50 in the EU. The AI Promotion Act (Act on the Promotion of Research, Development and Utilization of AI-Related Technologies) is a promotion law imposing duties to endeavor on businesses using AI, and I think it does not establish machine-readable marking of generated content or disclosure of deep fakes as legal obligations. Accordingly, for Japanese companies that do not provide services to the EU, the obligations under Article 50 do not immediately extend to their domestic business, but companies with EU-facing business need to manage the EU labeling requirements separately from their domestic generative AI use policy. The extent of liability where deep fakes or impersonation are used in advertising is also covered in Deepfake and Impersonation Advertising and PR Legal Affairs.

The specific items legal staff should check are as follows. First, check whether the users of the company's AI services, or the use of their output, extend to the EU, from both the contractual service area and actual access. Next, where this is the case, determine whether the company is a provider or a deployer under Article 50, or holds both positions. Further, if signing the code is being considered, settle, in view of the July 22 submission deadline, which officer within the company will sign as the "senior executive with sufficient authority." For the initial check items when a company launching an AI service has expansion in the EU in view, see The First Legal Checklist a Company Starting an AI Service Should Create, and for perspectives on incorporating EU-facing measures into an internal generative AI use policy, see How to Create a Generative AI Use Policy. For a cross-cutting check of the relationship with Japan's AI Promotion Act and the AI Guidelines for Business, Checkpoints for Responding to the AI Promotion Act and the AI Guidelines for Business may also be helpful.

Developments to Watch

As of the date of writing, the following points remain unsettled or awaiting future publication. The initial list of signatories is scheduled to be published before August 2, following the July 22, 2026 submission deadline, and whether Japanese companies' business partners and competitors sign will become clear at that point. The European Commission's implementing guidelines for Article 50 are said to have had a second draft published as of March 2026, with a final version planned before the August 2 date of application, and their final content, including the relationship with the code, has not been settled. In addition, the AI Office is said to consider formal updates to the code at least every two years, and future revisions may change the scope and the details of marking methods. As soon as these materials are published, the next task toward the August 2 date of application will be to check whether the company is among the signatories and to identify the differences from the company's obligations under Article 50.