Tentative Analysis of Principle Code for Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI (provisional title) (draft) Summary of Disclosure Items / Specific Examples

The 12th review meeting on Intellectual Property Rights in the Age of AI was held on April 23, 2026, and the following documents were attached to the meeting: “Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft)” (Reference Material 1-1), “Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft) Summary of disclosure items Specific examples” (Reference Material 2-1), “Major Opinions received through public comments (Intellectual Property Strategy Promotion Office, April 21, 2026)” (Material 1), etc.

Source: 内閣官房ホームページ(Cabinet Office Website)
・URLs of the review meeting:
https://www.cas.go.jp/jp/seisakukaigi/titeki2/ai_kentoukai/gijisidai/dai12/index.html
・URL of the Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft)” (Reference Material 1-1) :
https://www.cas.go.jp/jp/seisakukaigi/titeki2/ai_kentoukai/gijisidai/dai12/sanko1-1.pdf
・URL of the Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft) Summary of disclosure items / Specific examples” (Reference Material 2-1) :
https://www.cas.go.jp/jp/seisakukaigi/titeki2/ai_kentoukai/gijisidai/dai12/sanko2-1.pdf
・URL of the English version: “Principle-Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft)”  (Reference Material 1-2):
https://www.cas.go.jp/jp/seisakukaigi/titeki2/ai_kentoukai/gijisidai/dai12/sanko1-2.pdf
・URL of the English version: “Principles-Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft) Summary of disclosure items Specific examples” (Reference Material 2-2):
https://www.cas.go.jp/jp/seisakukaigi/titeki2/ai_kentoukai/gijisidai/dai12/sanko2-2.pdf

The above “Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title)” (hereinafter referred to as the “Proposed Principles Code”) shows, in “(2) Subjects to which this document is applied”, that even if a generative AI business does not have its head office or main office in Japan, the Proposed Principles Code shall apply if the generative AI system or generative AI service is provided to Japan (including, but not limited to, when it is available to Japanese nationals).

Moreover, the document states, “… regardless of purpose or whether they are a corporation or an individual”. Accordingly, the current draft shows that the “Proposed Principles Code” shall apply not only to large corporations, but also to non-profits, small companies, and individuals.

The Proposed Principles Code, if implemented as is, can have a significant impact on a wide range of companies and individuals even outside of Japan.  Many public comments have been submitted, and opinions have been presented from various perspectives.

However, the various positions reflect significant conflicts of interest, and a consensus on a satisfactory solution for all parties has not yet been reached. It is important to consider solutions that are satisfactory to all parties, including intellectual property right holders, developers of generative AIs, users of generative AIs, and the general public.

The current proposal has many problems. However, the majority of the opinions in the public comments do not necessarily advocate that the plan should be abolished. It is necessary to prepare for potential risks based on the current situation, even if revisions are considered in the future.  For example, a contract may need to be reviewed to prevent the leakage of contractual information from a counterparty or may be amended to include additional provisions to disclose information in the Principles Code.

The Proposed Principles Code may be misunderstood by some to be unimportant because it is a non-compulsory disclosure. However, it is risky to underestimate it, for example, because it can be a useful source of evidence for right holders to file lawsuits with businesses and for initial screening in the selection process.

In addition, it is necessary to fully recognize that this disclosure has the potential to lead to a wide variety of legal effects under Japanese laws, including claims for damages.

This article provides provisional comments on the “Principle Code for Protection of intellectual property and transparency for the appropriate use of generative AI (provisional title) (draft)  Summary of disclosure items / Specific examples” (Reference Material 2-1)” (hereinafter referred to as “Specific Examples”).

1.  Regarding the “Specific Examples”

(1) “Usage model related” (Table on Pages 1-2)

The “Specific Examples” disclose a version number. If many updates are made, modifications can be burdensome, especially to small companies and individuals. In addition, when there are many generative AIs, disclosure can be needed for each of them, and even for large enterprises, there is concern that the disclosure work will be burdensome. The release date and the modification date are also disclosed, and the burden may be increased if there are many modifications of functions.

Further, the “Specific Examples” disclose architecture/design specifications, the status of license contracts with third parties for model development, hardware and software required for use, license policy for users, terms of use, and details of the model training process.

Overall, there is a great burden in disclosing information, which can have stifling effects on small companies and individuals. For large enterprises, it may be necessary to consider an internal system where there is close cooperation between the technical division of generative AI and the legal and intellectual property divisions, etc. If it is not disclosed, it is necessary to explain the reason for such non-disclosure, which can become a difficult judgment. Since it is burdensome especially for foreign companies, they may require some support from Japanese law firms, etc.

(2) “Learning data” (Table on page 3)

The “Specific Examples” disclose items related to data used for learning and verification, etc. There will be positive aspects for disclosure, if it is possible to disclose that learning is done with clean training data.

Learning data, however, may include data that copyright holders hold the rights to, such as data collected through web crawling. In such cases, copyright holders will have an opportunity to obtain information about the training data and will be able to use it when considering legal action.

The “Specific Examples” also show the disclosure of information about web crawlers. In the future, it would be desirable to consider clean datasets to train generative AIs rather than use data collected through web crawling. However, the challenge lies in whether it is possible to obtain a volume of data comparable to what can be gathered through web crawling.

To address this challenge, the author proposed the realization of three key initiatives: the Data Income (DI) system, the Data Road Plan, and the Data Shinkansen Plan. These frameworks are designed to facilitate the large-scale collection of high-quality clean datasets.  It is crucial for national and local governments, corporations, and various organizations to collaborate in creating a global ecosystem of clean, large-scale datasets, supported by the aforementioned infrastructure plans. There may be a misconception that private data contracts alone are sufficient, but this is as misguided as assuming that nationwide road infrastructure could be built solely through privately funded toll roads.

(3) Accountability Relationships (Table on page 3)

The “Specific Examples” also disclose traceability, clearance of responsible people, distribution of responsibility among parties involved, etc. It shows disclosure items that can be used to pursue responsible people. There are risks of infringing intellectual property rights, including the possibility of shareholder derivative suits. It will be important for large companies to carefully consider what information they disclose.

The Proposed Principles Code, if implemented, would constitute a form of disclosure that may lead to various legal consequences in Japan, and therefore requires careful consideration.

(4) Measures to protect intellectual property rights (Table on pages 4-5)

The “Specific Examples” also disclose establishing a set of principles for protecting intellectual property rights.  Moreover, the “Specific Examples” disclose establishing a process to ensure that development and learning do not infringe on the intellectual property rights of other companies.

They also disclose compliance with paywalls and the use of web crawlers that comply with ‘robots.txt’. Careful attention will be required when using web crawlers and, in the future, it would be desirable to collect clean data sets rather than using web crawlers. In addition, they disclose keeping logs of learning for a certain period of time. This would be particularly burdensome to individuals.

It is necessary to avoid crawling of so-called pirated websites. This is ethical, but in some cases, it may be difficult to distinguish whether sites are pirated sites, and the use of crawlers can be risky because of the risk of disclosure requests from rights holders. If the Proposed Principles Code is implemented, an alternative should be provided in the form of clean, large-scale datasets.

The “Specific Examples” also mention filtering functions to prevent generated outputs infringing intellectual property rights. In order to enable the use of powerful filters by small/medium-sized companies and individuals without development capabilities, it is desirable to collect vast amounts of data to improve filters through the Data Income system and certify them as standard anti-infringement filters by the national government (Compliance Architecture). Disclosure of the use of certified “anti-infringement filters” should exempt other cumbersome disclosures. Implementation of the Proposed Principles Code without such an approach could seriously undermine the AI industry.

The “Specific Examples” also disclose C2PA compatible and electronic watermarking technology. With the Data Income System, Data Road Plan and Data Shinkansen Plan, it is important to create clean and transparent datasets.

The “Specific Examples” also disclose including clauses in the terms of use to prevent infringement of intellectual property rights. In the era of autonomous AI, such as AI Agents and AI employees, it is desirable for products per se to be guaranteed not to infringe on intellectual property rights. For instance, in the future, AI agents and AI employees will be able to make autonomous judgments on the spot and create and provide generated presentations to customers.  To address such future, in order to conduct such sales activities safely, AI should be trained with clean data sets or have a strong certified “anti-infringement filter”.

The current Proposed Principles Code is based on the old concept whereby a user must check when using the output of the generative AI. However, the era of autonomous AI is emerging, and most productivity gains will be lost if people must check the output of autonomous AI one step at a time. If the Proposed Principles Code is implemented as it is, there is concern that it will lead to future industry destruction.

The “Specific Examples” also disclose the establishment of copyright contact points and keeping correspondence records. Establishing and operating copyright contact points would place a heavy burden on individuals and small companies. There should be a system in place whereby an entity is exempted from maintaining a copyright contact point, if its clean datasets are certified by a national or other public authority and if it displays a “certified mark” indicating that learning has been conducted using certified data.

2. Tentative analysis of the “Specific Examples”

Large companies may be able to gather the wisdom of legal departments, intellectual property departments, and AI technical departments to properly prepare the disclosure based on the “Specific Examples”. However, careful consideration of disclosure will be crucial. Even for large companies, responding appropriately would not only be cumbersome and expensive, but also place a heavy burden on legal liability due to misdisclosure and disclosure of violations of Japanese laws and regulations.

These burdens differ from company to company, and in Japan there appear to be some companies that consider themselves capable of handling them, and others that consider themselves to be burdensome.

Japanese industry associations may choose to agree on the Proposed Principles Code by majority vote with proposals of revisions, and it may officially be introduced on the grounds that the industry is in favor of the Proposed Principles Code. However, if, for example, 20% of companies were to stop providing generative AI products or services because the burden is too heavy, this would be detrimental to the development of the AI industry.

Opinions of startup companies have been submitted, but what should be protected in the future includes not only start-up companies but also individuals who are considering starting a business. One serious concern is that the Proposed Principles Code would discourage entrepreneurship.

Also, even if 80% of companies think that the Proposed Principles Code is manageable, the age of autonomous AI may go beyond human imagination. Companies may later regret that the Proposed Principles Code has become an obstacle to new businesses questioning why they agreed to its introduction in the first place.

In order to prevent this from happening, it is important to create clean datasets. The Data Income system, the Data Road Plan, and the Data Shinkansen Plan need to be examined thoroughly.

In addition, it would be desirable for the government to collect a vast amount of data to improve the performance of the filtering system and to certify such systems as a standard “anti-infringement filter” (Compliance Architecture) enabling anyone to use a standardized filtering mechanism to prevent outputs that infringe intellectual property rights.

However, if only the Proposed Principles Code is implemented without having clean datasets in place, the impact on entrepreneurship will be significant and could have negative consequences for the industry.

Without clean, large-scale datasets, an entity may have to use web-crawled data to improve performance, and the copyright holders would not be able to benefit from the Proposed Principles Code. Even if disclosures under the Proposed Principles Code make it easier to collect evidence for litigation, the burden of pursuing litigation remains substantial, and the interests of copyright holders will still not be sufficiently protected.

If the Data Income system is implemented as a legal system, copyright holders will be able to choose whether to enter their work in the clean datasets to obtain Data Income or not (Opt-In).

The Data Income, Data Road Plan, and Data Shinkansen Plan can provide copyright holders with the choices of whether or not to allow AI to learn their work and provide a mechanism for obtaining remedies without filing a lawsuit.

3.  Summary

The current Proposed Principles Code, if implemented as is, will impose a heavy burden on a wide range of entities around the world, especially small companies and individuals.  Some entities, including large companies, who cannot bear the burden may abandon the development and provision of generative AI products or services in Japan by stifling effect, leading to a decline in the AI industry.

The current Proposed Principles Code requires reconsideration. If copyright protection is considered, the focus should be on developing clean datasets that are beneficial to everyone in the world, rather than implementing the Proposed Principles Code.

Since there is no clean database beyond the scale of web-crawled datasets, many generative AIs are trained from copyrighted, opaque data, such as data crawled from the Internet. This is the root cause, and what needs to be resolved is not to impose burdensome disclosures on companies and individuals, such as the Proposed Principles Code.

It is necessary to create a clean, large-scale database, remove non-clean copyrighted data, and establish a system of exemption and compensation by insurance, etc. in the case of using the database, and create a generative AI that ensures that the output does not infringe copyrights. In addition, it is desirable to establish a legal system that collects data on vast social norms and creates “anti-infringement filters” through the Data Income system so that the Compliance Architecture can be adopted and certified globally.

The age of autonomous AI, such as AI agents and AI employees, has already begun. Autonomous AIs, for example, can support sales, legal, intellectual property, general affairs, accounting, hiring, and education as employees and agents. Copyright problems should be completely resolved.

In the era of autonomous AI, including AI agents and AI employees, the creation of a legal infrastructure that can safely operate autonomous AI can dramatically increase productivity. To realize the “ultimate solution of generative AI and copyright issues”, it is necessary to discuss the construction of the infrastructure necessary for the autonomous AI era, such as the Data Income system, the Data Road Plan, the Data Shinkansen Plan, etc. In the era of advanced autonomous AIs, it is essential to ultimately resolve the “generative AI and copyright issues” and to prevent a substantial loss of GDP caused solely by the effects of copyright concerns.

In the era of autonomous AI, “ultimate solutions of generative AI and copyright issues” are necessary. Another ultimate solution is the approach being considered in India. This approach appears to be a harsh one for copyright owners: compulsory licensing of copyright. It may be considered to combine Opt-Out by copyright owners with this approach. However, the Data Income system (the Opt-In method) allows for the collection of overwhelmingly large data other than copyrighted materials.

In any case, there is an urgent need to create a clean, large-scale database. It will protect the interests of the copyright holders, developers, operators and users of generative AI, and people all over the world.

The creation of such database is not merely a technical challenge but will also lead to the collection of social norms through democratic process. Collecting data on social norms is also essential for safe and reliable actions of autonomous AI.

If large, clean datasets are created and appropriate legal systems are established for certifying such datasets, disclosure could be simplified to a single statement stating that “this AI is trained on the certified datasets, and the legal compliance of its output is guaranteed by law”. This would eliminate the need for any complex disclosures such as those required under the Proposed Principles Code.

Then, by developing legally safe sovereign AIs under the Data Income system, distilling to make “Democratic Local Sovereign AIs”, and distributing them to the public, everyone will be able to use copyright-safe AI, and the issue of “generative AI and copyrights” could ultimately be resolved.

Also, a registration system of a legally secure “certified AI” can also be created that is trained from a very large database collected by the Data Income system and that adopts Compliance Architecture. In this case, AI providers need only state “This AI is certified AI (registration number is XX”).”

In addition, by using a certified AI as a sovereign AI, the system of “Certified Sovereign AI” can be realized. The system of the “Certified Sovereign AI” enables the construction of infrastructure for the safe use of advanced autonomous AIs in society.

These are provisional comments on the examples of the Proposed Principles Code.  At this point in time in practice, it is important to consider potential risks if the Proposed Principles Code is implemented.

If the Proposed Principles Code is implemented as it is, it could have a large negative effect on the future of the world. It is important to pay attention to future discussions regarding the Proposed Principles Code.