« SEC Cybersecurity の開示昨年からの比較... | Main | オランダ データ保護庁 アルゴリズムに基づく意思決定における意味のある人間の介入 (2025.07.23) »

2025.07.31

欧州委員会 AI法における汎用AIモデル提供者の義務範囲に関するガイドライン(2025.07.18)

こんにちは、丸山満彦です、

欧州委員会が、AI法におけえる汎用AIモデル提供者の義務範囲に関するガイドラインを公表していますね...

2025.07.10に公表された、汎用AI実践規範(第1章「透明性」と第2章「著作権」第3章「安全とセキュリティ」)を補完するものという位置付けのようです...

 

 

⚫︎ European Commission

・2025.07.18 Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act

・・[PDF] Communication - Approval of the content of the draft Guidelines on the scope of the obligations for general-purpose AI model

COMMUNICATION TO THE COMMISSION  委員会へのコミュニケーション
Approval of the content of the draft Communication from the Commission – Guidelines on the scope of the obligations for general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act)  委員会からのコミュニケーションのドラフトの内容の承認 – 規則 (EU) 2024/1689 (AI 法) によって確立された汎用 AI モデルの義務の範囲に関するガイドライン
Regulation (EU) 2024/1689 of the European Parliament and the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending certain regulations (‘AI Act’) entered into force on 1 August 2024. Its aim is to promote innovation in and uptake of artificial intelligence (‘AI’) in the Union, while ensuring a high level of protection of health, safety and fundamental rights, including democracy and the rule of law.   欧州議会および理事会の規則(EU)2024/1689(2024年6月13日)は、人工知能に関する調和された規則を定めるとともに、特定の規則を改正する(『AI法』)もので、2024年8月1日に施行される。その目的は、健康、安全、および民主主義や法の支配を含む基本的権利の高度な保護を確保しつつ、欧州連合(EU)における人工知能(AI)のイノベーションと普及を促進することである。
Chapter V of the AI Act lays down obligations for the providers of general-purpose AI systems, divided into two categories: general-purpose AI model with and without systemic risk.   AI 法の第 5 章では、汎用 AI システムのプロバイダに対する義務が、システムリスクのある汎用 AI モデルとシステムリスクのない汎用 AI モデルという 2 つのカテゴリーに分類されて規定されている。
Pursuant to Article 96(1) of the AI Act, the Commission shall develop guidelines on the practical implementation of the AI Act. The draft Guidelines annexed to this Communication aim to increase legal clarity and to provide insights into the Commission’s interpretation of the provisions regarding general-purpose AI systems, in light of their imminent entry into application on from 2 August 2025.    AI 法第 96 条 (1) に基づき、欧州委員会は AI 法の実際的な実施に関するガイドラインを策定することになっている。このコミュニケーションに添付されているガイドラインのドラフトは、2025 年 8 月 2 日からの適用開始を控え、法的明確性を高め、汎用 AI システムに関する規定の欧州委員会の解釈について理解を深めることを目的としている。  
The draft Guidelines on general-purpose AI aim to clarify key concepts of the AI Act, such as what is a ‘general-purpose AI model’, a ‘provider of a general-purpose AI model’, a ‘placing on the market of a general-purpose AI model’, and how to estimate the computational resources used for training a general-purpose AI model. Beyond these conceptual clarifications, the draft Guidelines aim to clarify how the AI Office will support and work with providers who must comply with the general-purpose AI rules. In all above-mentioned cases, the proposed approach is pragmatic and practical.   汎用 AI に関するガイドライン草案は、AI 法における「汎用 AI モデル」とは何か、「汎用 AI モデルのプロバイダ」とは何か、「汎用 AI モデルの上市」とは何か、汎用 AI モデルのトレーニングに使用される計算リソースをどのように推定するかなど、AI 法の重要な概念を明確にするものである。これらの概念の明確化に加え、ガイドライン草案は、AI 事務局が、汎用 AI 規則を遵守しなければならないプロバイダをどのように支援し、連携していくかを明確にするものである。上記のすべての場合において、提案されているアプローチは実用的かつ実践的なものである。
The draft Guidelines are also part of the package that prepares the entry of application of the rules for providers of general-purpose AI models. The package will contain these draft Guidelines, the General-Purpose AI Code of Practice and the adequacy assessment by the Commission and the AI Board, the Template for transparency on training data, and further elements like a template for notifications that providers of general-purpose AI models with systemic risk to submit to the AI Office.   ガイドライン草案は、汎用 AI モデルプロバイダに対する規則の適用開始に備えるパッケージの一部でもあります。このパッケージには、本ガイドライン草案、汎用 AI 行動規範、欧州委員会および AI 委員会による妥当性評価、トレーニングデータの透明性に関するテンプレート、およびシステムリスクを伴う汎用 AI モデルの提供者が AI 事務局に提出する通知のテンプレートなどの追加要素が含まれる。
Such clarity is essential for providers of general-purpose AI models, since the obligations are directly applicable as from 2 August 2025. In view of this imminent entry into application and the time necessary for the translation of the guidelines into all languages, the Commission is called upon to approve the content of draft Communication from the Commission on Guidelines on the scope of the obligations for general-purpose AI models established by Regulation (EU) 2024/1689.  2025 年 8 月 2 日から義務が直接適用されるため、このような明確化は、汎用 AI モデルの提供者にとって不可欠である。この適用開始が間近であり、ガイドラインをすべての言語に翻訳するのに必要な時間を考慮して、委員会は、規則(EU)2024/1689 で定められた汎用 AI モデルの義務の範囲に関するガイドラインに関する委員会からのコミュニケーションのドラフトの内容を承認するよう求められている。
The Guidelines will be formally adopted by the Commission at a later date, when all language versions are available. It is only from that moment that these Guidelines will be applicable.  ガイドラインは、すべての言語版が利用可能になった時点で、後日、欧州委員会によって正式に採択される。このガイドラインは、その時点から適用される。
The text of the draft Guidelines is enclosed as an Annex to this Communication. ガイドラインのドラフトは、このコミュニケーションの附属書として添付されている。

 

・・[PDF] Guidelines on the scope of the obligations for general-purpose AI models established by AI Act

20250730-223411

・[DOCX][PDF] 仮訳

 

目次...

1. Background and objectives of the Commission guidelines 1. 委員会ガイドラインの背景と目的
2. General-purpose AI models 2. 汎用AIモデル
2.1. When is a model a general-purpose AI model? 2.1. モデルが汎用AIモデルである場合とは?
2.2. The lifecycle of a general-purpose AI model 2.2. 汎用AIモデルのライフサイクル
2.3. When is a general-purpose AI model a general-purpose AI model with systemic risk? 2.3. 汎用AIモデルがシステム的リスクを有する汎用AIモデルとなるのはいつであるか?
2.3.1. Classification 2.3.1. 分類
2.3.2. Notification 2.3.2. 通知
2.3.3. Procedure for contesting classification 2.3.3. 分類の異議申し立て手続き
2.3.4. Designation 2.3.4. 指定
2.3.5. Procedure for contesting designation 2.3.5. 指定の異議申立て手続
3. Providers placing on the market general-purpose AI models 3. 汎用AIモデルを市場に提供する事業者
3.1. When is an actor a provider placing on the market a general-purpose AI model? 3.1. いつ、行為者が汎用AIモデルの市場に供給する供給者となるか?
3.1.1. Examples of providers of general-purpose AI models 3.1.1. 汎用AIモデルの提供者の例
3.1.2. Examples of placing on the market of general-purpose AI models 3.1.2. 汎用AIモデルの市場への提供の例
3.1.3. Considerations for general-purpose AI models integrated into AI systems 3.1.3. AIシステムに組み込まれた汎用AIモデルに関する考慮事項
3.2. Downstream modifiers as providers of general-purpose AI models 3.2. 下流の改変者が汎用AIモデルの提供者としての地位
3.2.1. Downstream modifiers becoming providers of general-purpose AI models 3.2.1. 下流の修正者が汎用AIモデルの の提供者となる場合
3.2.2. Downstream modifiers becoming providers of general-purpose AI models with systemic risk 3.2.2. システム的リスクを有する汎用AIモデルの改変者となる下流の改変者
4. Exemptions from certain obligations for certain models released as open-source 4. オープンソースとして公開される特定のモデルに対する特定の義務の免除
4.1. Scope of the exemptions 4.1. 免除の適用範囲
4.2. Conditions for the exemptions to apply 4.2. 免除の適用条件
4.2.1. Conditions on the licence 4.2.1. ライセンスに関する条件
4.2.2. Lack of monetisation 4.2.2. 収益化の不存在
4.2.3. Public availability of parameters, including the weights, the information on the model architecture, and the information on model usage 4.2.3. パラメーターの公開性(重み、モデルアーキテクチャに関する情報、およびモデルの使用に関する情報)
5. Enforcement of the obligations for providers of general-purpose AI models 5. 汎用AIモデル提供者の義務の履行
5.1. Effects of adhering to a code of practice assessed as adequate 5.1. 適切なものと評価された行動規範の遵守の効果
5.2. Supervision, investigation, enforcement, and monitoring of Chapter V of the AI Act 5.2. AI法 第5章の監督、調査、執行、および監視
5.3. Compliance after entry into application of Chapter V of the AI Act on 2 August 2025 5.3. 2025年8月2日にAI法第5章が適用開始された後の遵守
6. Review and update of the Commission guidelines 6. 委員会ガイドラインの見直しおよび更新
Annex: Training compute of general-purpose AI models 附属書:汎用AIモデルの訓練計算
A.1. What should be estimated? A.1. 何を推定すべきか?
A.2. How should training compute be estimated? A.2. 訓練計算リソースはどのように見積もるべきか?
A.2.1. Hardware-based approach A.2.1. ハードウェアベースのアプローチ
A.2.2. Architecture-based approach A.2.2. アーキテクチャに基づくアプローチ
A.3. Training compute of models with approximately one billion parameters A.3. モデルパラメーターが約10億のモデルの訓練計算量

 

 


 

発表...

・2025.07.18 Learn more about the guidelines for providers of general-purpose AI models

Learn more about the guidelines for providers of general-purpose AI models 汎用 AI モデル提供者に関するガイドラインの詳細
The Commission published guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act. 欧州委員会は、AI 法に基づく汎用 AI モデル提供者の義務の範囲に関するガイドラインを発表した。
The aim is to provide legal certainty to actors across the AI value chain by clarifying when and how they are required to comply with these obligations. その目的は、AI バリューチェーンに関わる関係者に、これらの義務をいつ、どのように遵守すべきかを明確にし、法的確実性を提供することでである。
These guidelines are part of a broader package tied to the entry into application on 2 August 2025 of the EU-wide rules for providers of general-purpose AI models. They complement the General-Purpose AI Code of Practice that the Commission received from independent experts on 10 July. このガイドラインは、2025年8月2日に施行される、汎用 AI モデル提供者に関する EU 全体の規則と連動した、より広範なパッケージの一部である。これは、7月10日に独立専門家から欧州委員会に提出された「汎用 AI 行動規範」を補完するものである。
General-purpose AI models play a significant role in encouraging innovation and uptake of AI within the EU, as they can be used for a variety of tasks and integrated into a wide array of AI systems. For this reason, providers of such models have certain obligations under the AI Act. 汎用 AI モデルは、さまざまなタスクに使用でき、幅広い AI システムに統合できるため、EU における AI のイノベーションと普及を促進する上で重要な役割を果たしている。このため、このようなモデルの提供者は、AI 法に基づき一定の義務を負っている。
These obligations include giving information to providers of AI systems intending to integrate the model into their AI systems and putting in place a policy to comply with European copyright law. In addition, providers of the most advanced or most impactful general-purpose AI models, namely those presenting systemic risks, are subject to the additional obligations to assess and mitigate these systemic risks. Systemic risks include risks to fundamental rights and safety, and risks related to loss of control over the model. これらの義務には、モデルを AI システムに統合しようとする AI システム提供者に情報を提供すること、および欧州の著作権法を遵守するためのポリシーを策定することが含まれる。さらに、最も先進的または最も影響力の大きい汎用 AI モデル、すなわち、システミックリスクをもたらす AI モデル提供者は、これらのシステミックリスクを評価し、緩和する追加の義務を負う。システミックリスクには、基本的権利および安全に対するリスク、およびモデルに対する制御の喪失に関連するリスクが含まれる。
These guidelines clarify the scope of the obligations and to whom they apply. The guidelines focus on four key topics: 本ガイドラインは、義務の範囲および義務の対象者を明確にしています。本ガイドラインは、4 つの主要トピックに焦点を当てている。
・General-purpose AI models:  An AI model is considered to be a general-purpose AI model if it was trained using an amount of computational resources (‘compute’) that exceeds 10^23 floating point operations and if it can generate language (whether in the form of text or audio), text-to-image or text-to-video. ・汎用 AI モデル:AI モデルは、10^23 以上の浮動小数点演算の計算リソース(「演算能力」)を使用して訓練され、言語(テキストまたは音声の形式を問わず)、テキストから画像、またはテキストから動画を生成できる場合、汎用 AI モデルとみなされる。
・Providers of general-purpose AI models: The guidelines outline the concepts of a ‘provider’ and of ‘placing on the market’ and clarify when an actor modifying a general-purpose AI model is considered to become a provider. ・汎用 AI モデル提供者:このガイドラインでは、「提供者」および「上市」の概念の概要を説明し、汎用 AI モデルを改変する事業者が提供者とみなされる場合について明確化している。
・Exemptions from certain obligations: The guidelines clarify under what conditions providers of general-purpose AI models released under a free and open-source license and satisfying certain transparency conditions may be exempt from certain obligations under the AI Act. ・特定の義務の免除:このガイドラインでは、フリーでオープンソースのライセンスに基づいてリリースされ、特定の透明性条件を満たす汎用 AI モデル提供者が、AI 法に基づく特定の義務を免除される条件について明確化している。
・Enforcement of obligations: The guidelines explain the implications for providers of general-purpose AI models that choose to adhere and implement the General-Purpose AI Code of Practice, and outline Commission expectations regarding compliance as from 2 August 2025. ・義務の執行:このガイドラインでは、汎用 AI コード・オブ・プラクティスを遵守し、実施することを選択した汎用 AI モデル提供者に対する影響について説明し、2025 年 8 月 2 日以降の遵守に関する欧州委員会の期待事項の概要を記載している。
Next steps 次のステップ
The AI Act obligations for providers of general-purpose AI models enter into application on 2 August 2025. From that date onwards, providers placing general-purpose AI models on the market must comply with their respective AI Act obligations. Providers of general-purpose AI models that will be classified as general-purpose AI models with systemic risk must notify the AI Office without delay. In the first year after entry into application of these obligations, the AI Office will work closely with providers, in particular those who adhere to the General-Purpose AI Code of Practice, to help them comply with the rules. From 2 August 2026, the Commission’s enforcement powers enter into application. 汎用 AI モデル提供者に対する AI 法の義務は、2025 年 8 月 2 日から適用されます。この日から、汎用 AI モデルを上市する提供者は、AI 法に基づくそれぞれの義務を遵守しなければならない。システムリスクのある汎用 AI モデルに分類される汎用 AI モデル提供者は、AI 事務局に遅滞なく通知しなければならない。これらの義務の適用開始後 1 年間は、AI 事務局は、特に汎用 AI 行動規範を遵守する提供者と緊密に連携し、規則の遵守を支援する。2026年8月2日より、欧州委員会の執行権限が適用される。
Providers of general-purpose AI models already on the market before 2 August 2025 must comply with the relevant obligations under the AI Act by 2 August 2027. 2025年8月2日より前にすでに市場に投入されている汎用AIモデルの提供者は、2027年8月2日までにAI法に基づく関連義務を遵守しなければならない。
Background 背景
The AI Act’s obligations for providers of general-purpose AI models aim to ensure that general-purpose AI models are transparent, in line with EU and national copyright law, and that providers of the most advanced or most impactful general-purpose AI models assess and mitigate the systemic risks presented by these models. To help providers of general-purpose AI models comply with the AI Act, the Commission is making available a package of information, which includes the guidelines, the General-Purpose AI Code of Practice, and further tools to assist providers, such as the AI Act Service Desk. AI 法における汎用 AI モデルの提供者に対する義務は、汎用 AI モデルが EU および各国の著作権法に準拠し、透明性を確保するとともに、最先端または影響力の大きい汎用 AI モデル提供者が、これらのモデルがもたらすシステミックリスクを評価し、緩和することを目的としている。汎用 AI モデルの提供者が AI 法を遵守するのを支援するため、欧州委員会は、ガイドライン、汎用 AI 行動規範、AI 法サービスデスクなどの提供者を支援するためのツールを含む一連の情報を公開している。
The guidelines were developed through a public consultation, during which the Commission gathered input from hundreds of stakeholders. They also reflect input from the expert pool set up by the Commission’s Joint Research Centre to advise the AI Office on the classification of AI models, including general-purpose AI models and those with systemic risk, as well as feedback from other experts. このガイドラインは、委員会が数百人の利害関係者から意見を収集した公開協議を経て作成された。また、汎用 AI モデルやシステミックリスクのある AI モデルなどの AI モデルの分類について AI 事務局に助言を行う、委員会合同研究センターが設立した専門家プールからの意見や、その他の専門家からのフィードバックも反映されている。
While not legally binding, these guidelines set out the Commission’s interpretation and application of the AI Act, which will guide its enforcement actions. このガイドラインは法的拘束力はありませんが、AI 法に関する委員会の解釈と適用を規定しており、その施行措置の指針となる。
Find more information about the Guidelines for providers of general-purpose AI models: 汎用 AI モデル提供者向けガイドラインの詳細については、以下を参照。
Q&A Q&A
Webpage ウェブページ

 

 


 

参考

AI法に関する詳細な欧州委員会による説明

AI Act

 

欧州委員会意見書(EU2024/1689規則第56条に基づく汎用AI行動規範の評価

・2025.08.01 Commission Opinion on the assessment of the General-Purpose AI Code of Practice

・・[PDF] COMMISSION OPINION of 1.8.2025 on the assessment of the General-Purpose AI Code of Practice within the meaning of Article 56 of Regulation (EU) 2024/1689

20250902-52258

・[DOCX][PDF] 仮訳

 

 


 

⚫︎ まるちゃんの情報セキュリティ気まぐれ日記

・2025.07.25 欧州委員会 汎用 AI 実践規範 (2025.07.10)

・2025.04.11 欧州委員会 AI大陸行動計画 (2025.04.09)

・2025.03.13 欧州委員会 AI法における汎用AIモデル - Q&A (2025.03.10)

・2025.02.08 欧州委員会 規則(EU)2024/1689(AI法)が定める人工知能の禁止行為に関する欧州委員会ガイドライン

・2024.12.22 欧州委員会 汎用AI実践規範の第2ドラフトを発表 (2024.12.19)

・2024.12.06 欧州委員会 AI法における汎用AIモデル - Q&A (2024.11.20)

・2024.11.17 欧州委員会 汎用AI実践規範の最初のドラフトを発表 (2024.11.14)

・2024.10.30 欧州 AI法の調和標準の策定について...

・2024.09.08 欧州評議会 AI条約の署名開始... (2024.09.05)

・2024.08.05 欧州AI法が施行された... (2024.08.01)

・2024.07.19 ドイツ BfDI フランス CNIL オランダ AP EU AI法関連

・2024.07.16 EU 2024.07.12にAI法がEU官報に掲載された

・2024.05.22 EU 欧州理事会がAI法を承認...まもなく発効されますね...

・2024.03.19 欧州議会 AI法 (2024.03.13) とサイバーレジリエンス法を採択 (2024.03.12)

・2023.12.10 欧州委員会、欧州議会がAI法について政治的合意

・2022.12.08 EU理事会 AI法に関する見解を採択

・2022.09.30 欧州委員会 AI責任指令案

・2021.12.05 欧州理事会 AI法改正案を欧州議会に提出

・2021.08.08 EU議会 BRIEFING 人工知能法 at 2021.07.26

・2021.04.24 欧州委員会がAIへの規制を提案 → 欧州データ保護官は歓迎するけど、公共空間での遠隔生体認証についての規制も入れてね

 

|

« SEC Cybersecurity の開示昨年からの比較... | Main | オランダ データ保護庁 アルゴリズムに基づく意思決定における意味のある人間の介入 (2025.07.23) »

Comments

Post a comment



(Not displayed with comment.)


Comments are moderated, and will not appear on this weblog until the author has approved them.



« SEC Cybersecurity の開示昨年からの比較... | Main | オランダ データ保護庁 アルゴリズムに基づく意思決定における意味のある人間の介入 (2025.07.23) »