The 「生成ＡＩの適切な利活用等に向けた知的財産の保護及び透明性に関するプリンシプル・コード（仮称）（案）」 — Draft Principles Code on the Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI (provisional title) — is an 11-page Japanese-language consultation document released for public comment on the e-Gov portal by the Cabinet Office's Intellectual Property Strategy Headquarters Secretariat (内閣府知的財産戦略推進事務局) in August 2026 (Source: public-comment.e-gov.go.jp).
It is a draft, marked both 案 ("draft") and 仮称 ("provisional title"), carries no penalties, and is not in force. Its stated purpose is to set out principles for transparency and intellectual-property protection that generative-AI operators should take, so as to reconcile the advance of the technology with the protection of intellectual-property rights and secure a usage environment that rights holders and users can rely on.
The document names its own two models directly: the transparency and copyright measures of the EU AI Act, and comply-or-explain practice from corporate-governance stewardship codes. It states that it is drawn up under the purport of the Act on Promotion of Research and Development and Utilization of AI-Related Technology (令和七年法律第五十三号, Act No. 53 of 2025) — the AI Promotion Act.
No official English translation was located. Japanese terms are given below alongside a working rendering.
Mechanism: comply-or-explain
The document's governing choice is set out in section 1(3), and is the feature most easily lost in summary. The code 「強制的な開示を求めるものではなく」 — does not demand compulsory disclosure of information belonging to the operator, trade secrets expressly included — and instead asks operators to respond to each principle by コンプライ・オア・エクスプレイン, implementing it or explaining why not.
The explanation is not pro forma. An operator declining a principle must explain in a way calculated to secure the adequate understanding of users and rights holders. Two further constraints tighten it:
- The override clause is not an explanation. Section 2(3) states that pointing to a contract or terms-of-service clause removing or limiting the code's principles (an オーバーライド条項) is insufficient on its own; the operator must explain why that clause exists. An operator that has declared acceptance but narrows its disclosure through its terms of use is treated as needing a separate explanation.
- "Not yet built" is not an explanation. The commentary to Principles 2 and 3 states that merely saying the operator has not established the capacity to implement the principle is insufficient; the operator must give an appropriate account of when that capacity will be in place, weighing its own business scale.
Scope
The code applies to 生成ＡＩ事業者 (generative-AI operators), a term covering two roles:
- 生成ＡＩ開発者 (developers) — those who build a generative-AI system through model and algorithm development, data collection including purchase, preprocessing, training and validation, and who have provided all or part of that system to the public. Purpose and legal form are immaterial.
- 生成ＡＩ提供者 (providers) — those who validate, integrate, supply or operate a generative-AI service embedding such a system into an application, product, existing system or business process, and who have provided it to the public.
Two exclusions are stated for clarity: a party that supplies a generative-AI system used only by itself, on its own data, is not a developer under the code; and a party that supplies a service specialised on one entity's own data only to that entity is not a provider.
Scope reaches beyond Japan. An operator with neither head office nor principal office in Japan falls within the code where its system or service is offered toward Japan, a condition the text states includes — but is not limited to — cases where Japanese nationals can use it.
Registration and visibility
Section 1(4) asks accepting operators to publish, on their corporate website or an equivalent site, a declaration of acceptance (受入れ表明) together with, for each principle, what they implement or their reason for not implementing it; to file the same with the Secretariat on a prescribed form; and to review and update it annually, publishing the fact of an update.
The Secretariat's own role is bounded and stated in the same section. It prepares a reference form, publishes a list of filing operators and links to their disclosures, and works with relevant ministries and industry bodies to encourage filing. It 「当該届出の内容について審査を行うものではなく」 — does not review the contents of filings — and does not respond to third-party enquiries about them.
The three principles
Principle 1 — summary disclosure
Operators publish on a publicly viewable corporate site a summary of the following (概要開示対象事項), accessible to all including users and rights holders.
Transparency measures. Under model: name, including identifier and version; provenance including release date, past versions and revision history; architecture and design specification, including third-party licence arrangements in development and the hardware, software and licences needed for use; usage terms, clarifying intended uses and restricted or prohibited ones; and the content of the model's training process, including training method and parameter settings covering the inference process and basis of judgment. Under training data: matters relating to the data used for training and validation — data types, non-public datasets obtained by web crawl or from third parties, public datasets, data collected by other means, and whether and why synthetic data was used — and crawler details, covering purpose, collection period, name and identifier, and whether third-party crawlers were used and under what names. Under accountability: the state of traceability of decisions made during development, provision and use, so far as technically possible and reasonable, including designation of responsible parties, allocation of responsibility, concrete responses to stakeholders, and documentation.
Intellectual-property protection measures. Operators disclose their status on: formulating IP-protection principles with a clear responsibility structure, reviewing them at least annually and publishing a summary; not infringing others' intellectual-property rights in data use including development and training; respecting access restrictions such as paywalls and adopting crawlers that obey machine-readable instructions including robots.txt, publishing those measures per user-agent and giving notice on change; retaining training logs for a fixed period; avoiding crawling so-called pirate sites; taking technical measures so far as possible to prevent generation of infringing output; taking technical measures so far as possible for provenance, such as digital watermarking and C2PA; informing users that output believed to infringe should not be used; and maintaining an appropriate contact point for rights-holder claims, with requirements clarified so far as possible and records of responses kept.
A footnote grounds the log-retention item in the Ministry of Internal Affairs and Communications / METI AI Business Operator Guidelines (ＡＩ事業者ガイドライン, version 1.1), page 18, on securing verifiability.
Principle 2 — enquiry by rights holders in or preparing legal proceedings
A 原則２開示要求者 is a person actually conducting or preparing litigation, mediation, ADR or other legal procedure to realise their rights or legally protected interests, or a lawyer or statutory representative instructed by them. On request, and where the request meets all three conditions below, the operator answers:
- whether specified URL information the requester supplies (原則２対照情報, limited to what the operator can readily access and confirm) is contained in the data used for training and validation — including web-crawled data, non-public third-party datasets, public datasets, synthetic data, and data collected by other means; and
- where the recipient is a provider unable to answer, the name of the party that developed the model in the service.
The request must (1) state reasons showing the requester qualifies, (2) state the purpose for which the answer will be used and pledge not to use it otherwise, and (3) identify the comparison information and the grounds for seeking disclosure in relation to it.
The worked example given is a creator who publishes work on website A, finds generated output identical or similar to it, and asks whether that domain was within the crawler's scope or among the sources of third-party training data.
Principle 3 — enquiry by users of generated output
A 原則３開示要求者 is a person who has generated, using the operator's system or service, a film, music, drama, literary work, photograph, manga, animation, computer game, or other text, figure, colour, sound, motion or video, or a program for providing such information by computer. The enquiry mirrors Principle 2 but asks whether content identical or similar to the requester's own output — as defined by Article 2(1) of the Act on Promotion of Creation, Protection and Exploitation of Content — appears at a URL in the training data.
The conditions differ in two respects. The requester must supply both the generated output and the prompt used to generate it. And the pledge is broader: not to use the answer for any other purpose or for litigation, mediation, ADR or other legal procedure — the reverse of Principle 2's premise.
Common commentary
For both enquiry principles the code asks operators to disclose in as much detail and as intelligibly as possible; to consider and confer in good faith even where the matter may be a trade secret; and to publish their own response policy so as to avoid excessive burden. No response deadline is fixed, but prompt disclosure within a reasonable period is expected. Both note that other routes exist after suit is filed — inter-party enquiry under Article 163 of the Code of Civil Procedure and applications for document production orders under Article 221.
On abuse, both contemplate a fee or a cap on the number of requests within a given period, with the express caution not to set fees or caps that would deter, obstruct, or cause requesters to give up.
Open-source exception
Section 2(2) provides one exception covering all three principles. An operator conducting development or training — pre- or post-training — that uses open-source software for all or part of its business, and for which the use of that software makes some Principle 1 summary item impossible to either disclose or explain, or makes some Principle 2 or 3 item impossible to disclose, may substitute disclosure of the fact of open-source use and the details of the licence involved.
Incentives and revision
Section 2(4) states an expectation that government will evaluate operators' published content and concrete measures and provide some form of incentive within the schemes and programmes it operates. The code is to be revised where necessary in light of operator responses and international developments.
Matters still under consideration
A closing 【留意事項】 lists five open items: the substance of the Principle 1 summary-disclosure items and the Principle 2 and 3 enquiry items; the granularity of disclosure under each principle; how Principles 2 and 3 should relate to one another; what accommodation is owed to startups, given the code addresses those providing systems or services to the public; and how to spread the principles to operators within scope — including those without a Japanese head office — and what incentives to use.
Provenance and limits
Retrieved on August 25, 2026 from the canonical e-Gov public-comment portal URL, HTTP 200, all 11 pages, through a PDF parser; the host returns HTTP 403 to a plain command-line fetch. The document was independently verified and read in full on August 24, 2026.
The retrieved text carries no date of its own; the August 18, 2026 date of record comes from the coverage that surfaced the consultation, not from the document. The comment period's closing date, the number of comments received, and any subsequent revision are outside the retrieved text.
Relationships
- depends-on: Japan AI Promotion Act (Act on Promotion of Research, Development and Utilisation of AI-Related Technologies, 2025) — the code is drawn up under the purport of Act No. 53 of 2025.
- related: EU AI Act (Regulation 2024/1689) — named in the text as the model for its transparency and copyright measures.
- related: AI Copyright — the enquiry principles create a training-data disclosure route directed at copyright disputes.
- related: Data Provenance, C2PA, and Watermarking — Principle 1 asks for watermarking and C2PA-style provenance measures.
- related: Web Scraping for AI Training — the crawler-disclosure and robots.txt items, and both enquiry rights, turn on how the training corpus was gathered.
- related: EU vs. US AI Regulation: A Deep Comparison — a third regulatory model beside the EU and US approaches, borrowing EU substance through a soft-law mechanism.