Citoyen
  • ホーム
  • フィード

探す

  • 国
  • カテゴリ
  • 比較
  • フォロー中

理解する

  • パノラマ国別・カテゴリ別
  • 特集詳細な分析
  • 任期の総括各指導者の実績
  • 改革進捗とシミュレーター

データ

  • 出典
  • カレンダー
  • 方法
  • AIエージェントと開発者

Citoyen

  • 概要
  1. 欧州連合
  2. ›
  3. 人工知能
施行中

EU Artificial Intelligence Act (2024) and its 2026 « Digital Omnibus » amendment

Regulation (EU) 2024/1689, in force since 1 August 2024, regulates AI according to risk: eight practices banned since 2 February 2025, rules for general-purpose AI models since 2 August 2025, obligations for high-risk systems, transparency duties, fines of up to €35m or 7% of worldwide turnover. Regulation (EU) 2026/1744 (« Digital Omnibus on AI »), in force since 27 July 2026, postponed the high-risk obligations to 2 December 2027 and 2 August 2028 and added a ban on « nudification » apps and AI-generated child sexual abuse material from 2 December 2026.

Citoyen·更新日:2026年11月5日

この特集は、新しい数値を反映して、少なくとも四半期に1回更新されます。 フォローすると、更新のたびに通知を受け取れます。

要点

禁止されたAI行為に対する最高制裁金

7%

全世界の年間売上高に対する割合 · 発表値

規則第99条第3項による[1][2]

または3,500万ユーロのいずれか高い方。中小企業および小規模中堅企業については、低い方。

AI規則に賛成した欧州議会の票

523

2024年3月13日 · 観測値

欧州議会による[7]

反対46、棄権49。

2026年のオムニバスに賛成した欧州議会の票

423

2026年6月16日 · 観測値

欧州議会による[9]

反対57、棄権174。

AIに関するデジタル・オムニバスによる行政コスト削減見込み(上限)

0.43十億ユーロ

当初の推計、2025年11月 · 発表値

欧州委員会による(提案COM(2025) 836の理由書)[5]

当初提案についての約2億9,720万~4億3,320万ユーロの幅。影響評価は実施されていません。

  • 状況:施行中 2026年8月2日以降。 Applies from 2 August 2026, including the transparency duties of Article 50 and the Commission’s fines on general-purpose AI model providers; the high-risk requirements follow the dates set in 2026 (December 2027 / August 2028).[1][2][9]
  • 次の節目:2026年12月2日、「脱衣化」およびAI生成の児童性的虐待コンテンツの禁止が適用開始、すでに市場にある生成系システムの表示期限

概要

基礎となる文書

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)[1]

第2024/1689号 · 2024年6月13日

本文を読む ↗

推進主体

  • Thierry Breton[4]Member of the European Commission, leading the 2021 proposal (DG CONNECT) · 2021 · 発議者
  • Brando Benifei, Dragoş Tudorache[7]Co-rapporteurs of the European Parliament (Internal Market and Civil Liberties committees) · 2024 · 発議者
  • Henna Virkkunen[6]Member of the European Commission, leading the « Digital Omnibus on AI » proposal (DG CONNECT) · 2025 · 修正
  • Arba Kokalari, Michael McNamara[8][9]Co-rapporteurs of the European Parliament on the « Digital Omnibus on AI » (Internal Market and Civil Liberties committees) · 2026 · 修正

実施主体

  • European Commission, European AI Office (supervision of general-purpose AI models)
  • Member States, national market surveillance and notifying authorities
  • European Artificial Intelligence Board (representatives of the Member States)

掲げられた目的

域内市場の機能を改善し、人間中心で信頼できるAIの普及を促進するとともに、AIシステムの有害な影響から、健康、安全および基本的権利(民主主義、法の支配、環境保護を含む)を高い水準で保護し、イノベーションを支援すること。2026年の改正は、規格と当局の整備の遅れを踏まえ、「保護の水準を下げることなく」実施を円滑にすることを目的としています。[1][2]

規則(EU)2024/1689第1条および規則(EU)2026/1744の前文第2項・第3項による

主な措置

  • 8つの禁止行為(第5条):有害な操作または欺瞞、脆弱性の悪用、ソーシャルスコアリング、プロファイリングのみに基づく犯罪リスクの予測、顔画像の無差別な収集、職場と学校での感情認識(医療上または安全上の理由を除く)、機微な属性を推定する生体分類、そして法執行機関による公共空間でのリアルタイム遠隔生体識別(事前の許可を要する限定的な例外を除く)。[1][14]
  • 2026年に追加された9番目の禁止事項で、2026年12月2日から適用されます。特定可能な個人の写実的な私的画像、動画または音声を本人の同意なく生成するAIシステム(「ヌード化」)、または児童性的虐待コンテンツを生成するAIシステムが対象です。提供者については、それがシステムの目的である場合、または適切な安全措置がなければ予見可能な結果である場合、利用者については、その目的で使用する場合に適用されます。[2][9]
  • 高リスクシステム(EUの安全関連法令の対象製品の安全部品として使われるAI(附属書I)、および附属書IIIに列挙された8分野で使われるAI。例:採用、与信評価、教育、法執行、移民):リスク管理、データ品質、文書化、ログの記録、人間による監督、正確性と堅牢性、適合性評価が求められます。[1][7]
  • 汎用AIモデル:技術文書、下流の提供者向け情報、著作権遵守方針、学習用コンテンツの公開要約が求められます。10²⁵回を超える浮動小数点演算で学習したモデルはシステミックリスクがあると推定され、これに加えて(敵対的テストを含む)評価、リスクの軽減、重大インシデントの報告、サイバーセキュリティの確保が必要です。[1]
  • 透明性:AIシステムと対話していることを人々に知らせなければなりません。AIが生成した音声、画像、動画またはテキストは、機械可読形式で表示しなければなりません。ディープフェイクは、それを公開する者が明示しなければなりません。[1]
  • 加盟国が定める制裁:禁止された行為については最大3,500万ユーロまたは全世界の年間売上高の7%(いずれか高い方)、その他の大半の義務については1,500万ユーロまたは3%、誤解を招く情報については750万ユーロまたは1%です。中小企業(および2026年以降は小規模中堅企業)には、低い方の額が適用されます。欧州委員会は、汎用AIモデルの提供者に最大1,500万ユーロまたは3%の制裁金を科すことができます。[1][2]
  • ガバナンス:欧州委員会のAIオフィスが、汎用AIモデルに関する規則(情報提供要求、評価、制裁金)を執行し、同一の提供者のモデルに基づくAIシステムを監督します。この役割は2026年に強化されました。その他は各国の市場監視当局が執行します。AI委員会は加盟国で構成されます。[1][2][13]
  • 2026年の簡素化:「安全部品」の定義の限定、AIリテラシー義務を職員のリテラシー向上を支援する措置を講じる義務へ緩和、バイアスの検出と是正のための個人データの例外的な処理、機械類を固有の分野別安全規則へ移管、中小企業向けの負担軽減を小規模中堅企業へ拡大。[2][1]

データで見る

AIを利用する企業(従業員10人以上)

20%

2025

Eurostatカードを見る →

生成AIツールを利用する人

32.7%

2025

Eurostatカードを見る →

年表

  1. 2021年4月21日発表済み

    Commission proposal COM(2021) 206[3][4]

  2. 2023年6月14日審議中

    Parliament adopts its amendments: negotiations with the Council[4]

  3. 2023年12月

    Provisional agreement between Parliament and Council negotiators[7]

  4. 2024年1月24日

    Commission decision setting up the AI Office[1]

    The regulation defines the AI Office as the Commission’s function for implementing, monitoring and supervising AI systems and general-purpose AI models, provided for in the Commission Decision of 24 January 2024.

  5. 2024年3月13日

    European Parliament adopts its first-reading position: 523 for, 46 against, 49 abstentions[7][4]

  6. 2024年5月21日可決

    Council approves Parliament’s first-reading position: the act is adopted[4][1]

  7. 2024年6月13日

    Signature by the Presidents of the Parliament and of the Council[4][1]

  8. 2024年7月12日

    欧州連合官報に公布[1][4]

  9. 2024年8月1日施行中

    Entry into force (twentieth day after publication)[1][2]

    In force, with a staggered application: no obligation applies yet on that date.

  10. 2025年2月2日施行中

    Bans on prohibited AI practices and AI-literacy duty apply[1][14]

    Chapters I and II apply: general provisions (including AI literacy) and the eight prohibited practices of Article 5.

  11. 2025年7月10日

    汎用AI向け行動規範の公表[11][12]

    13人の独立した専門家が、1,000を超える利害関係者の意見を踏まえて作成した自主的な規範で、3章(透明性、著作権、安全性とセキュリティ)から成ります。欧州委員会とAI委員会は、適合性を示すうえで適切であると判断しました。欧州委員会が公表した署名者には、Amazon、Anthropic、Google、IBM、Microsoft、Mistral AI、OpenAIが含まれます。

  12. 2025年8月2日施行中

    Rules for general-purpose AI models, governance and penalties apply[1]

    Notified bodies (Chapter III, Section 4), general-purpose AI models (Chapter V), governance (Chapter VII), penalties (Chapter XII, except the Commission’s fines on model providers) and confidentiality apply.

  13. 2025年11月19日

    欧州委員会が「AIに関するデジタル・オムニバス」を提案(COM(2025) 836)[5][6]

    整合規格の遅れと各国当局の整備の遅れを理由に、この提案は、「高リスク」義務の適用開始を、適合対応のための手段が利用可能であることを確認する欧州委員会の決定にかからせ、期限を2027年12月2日と2028年8月2日としました。また、すでに市場にある生成系システムの提供者に対し、AI生成コンテンツの表示を行うまでの猶予を2027年2月2日まで認めました。

  14. 2026年3月13日

    理事会がAIに関するデジタル・オムニバスについての立場を決定[10]

  15. 2026年3月26日

    欧州議会がオムニバスに対する修正案を採択[6]

  16. 2026年5月7日

    オムニバスについて欧州議会と理事会が暫定合意[8]

    交渉担当者は「高リスク」義務について確定的な期日を定め、AI生成コンテンツの表示期限を、欧州委員会提案の2027年2月2日から2026年12月2日に前倒ししました。

  17. 2026年6月16日

    欧州議会がオムニバスを承認:賛成423、反対57、棄権174[9][6]

  18. 2026年6月29日

    理事会が欧州議会の立場を承認:オムニバスが成立[6][2]

  19. 2026年7月24日

    2026年7月8日の規則(EU)2026/1744が官報に掲載[2][6]

    2026年7月8日に署名され、公表の3日後に発効しました。

  20. 2026年7月27日施行中

    Amended by Regulation (EU) 2026/1744 (« Digital Omnibus on AI »)[2][6]

    High-risk obligations postponed to 2 December 2027 (Annex III uses) and 2 August 2028 (AI in products under Annex I legislation); new ban on « nudification » and AI child sexual abuse material from 2 December 2026.

  21. 2026年8月2日施行中

    General date of application: most remaining provisions apply[1][2][9]

    Applies from 2 August 2026, including the transparency duties of Article 50 and the Commission’s fines on general-purpose AI model providers; the high-risk requirements follow the dates set in 2026 (December 2027 / August 2028).

  22. 2026年12月2日今後の予定:結果は未定です

    「脱衣化」およびAI生成の児童性的虐待コンテンツの禁止が適用開始、すでに市場にある生成系システムの表示期限[2]

    第5条第1項の新たな(ba)号と(bb)号が適用されます。2026年8月2日より前に市場に投入された、合成の音声、画像、動画またはテキストを生成するシステムの提供者は、この時点までに、出力を機械可読かつ検出可能な形で表示しなければなりません(第50条第2項)。

  23. 2027年12月2日今後の予定:結果は未定です

    附属書IIIに掲げる高リスクAIシステムの義務が適用開始[2][1]

    生体認証、重要インフラ、教育、雇用、重要なサービスへのアクセス、法執行、移民・国境管理、司法および民主的プロセスが対象です。当初の期限は2026年8月2日で、2026年の改正前のものです。

  24. 2028年8月2日今後の予定:結果は未定です

    EUの安全関連法令の対象製品に組み込まれる高リスクAIの義務(附属書I)が適用開始[2][1]

    当初の期限は2027年8月2日で、2026年の改正前のものです。

議論の論点

Citoyenは特定の立場を取りません。主張は、発言者の表現のまま紹介しています。

支持する側の主張

  • 欧州議会の共同報告者は、これを世界初の拘束力あるAI法と位置づけ、リスクの低減、差別への対処、透明性の確保を目的として、容認できない行為を禁止し、労働者と市民の権利を保護するものだと説明しました(ブランド・ベニファイ、S&D)。またドラゴシュ・トゥドラケ(Renew)は、AIをEUの基本的価値と結びつけるものだと述べました。[7]
  • 欧州委員会と共同立法者によると、2026年の延期は、整合規格と各国当局の整備が遅れ、適合対応が想定より重くなったことへの対応です。欧州委員会は、節減額を2億9,700万ユーロから4億3,300万ユーロと見積もっていました。欧州経済社会委員会(EESC)は、「高リスク」の期限を規格の整備状況に合わせるよう勧告していました。[2][5][16]
  • オムニバスの共同報告者アルバ・コカラリ氏(EPP)は、「一時停止ボタンを押す」ことと事務手続の削減により、欧州でのテック企業の設立が容易になると主張しています。共同報告者のマイケル・マクナマラ氏(Renew)は、この文書は規則の構造を維持しつつ法的安定性をもたらすと述べ、さらにヌード化アプリの全面禁止を加えるとしています。[9]

反対する側の主張

  • 2021年、欧州経済社会委員会は禁止の範囲が狭すぎると判断しました。公共または私的にアクセス可能な空間での自動生体認識(認証を除く)と、行動シグナルの認識を禁止するよう求めました。しかし最終的な文書は、列挙された場合における警察によるリアルタイムの識別を認め、その他の生体認証の用途を高リスクに分類しています。[15][1]
  • 欧州地域委員会は、オムニバス・パッケージについて、基本的権利の保障への影響を測るための影響評価や公開協議が行われなかったことを遺憾としました。欧州委員会は、改正は技術的なものと判断し、影響評価は実施しなかったとしています。[17][5]
  • EESCは、従業員に十分な水準のAIリテラシーを確保する明示的な義務を維持することを求めました(2026年の文書では、これを支援するための措置を講じる義務に置き換えられました)。また、提供者が附属書IIIのシステムを高リスクではないと分類する場合に、当局に対する透明性を維持することも求めました。[16][2]

注目すべき指標

  • AIを利用する企業(従業員10人以上) · Eurostatカードを見る →
  • 生成AIツールを利用する人 · Eurostatカードを見る →
  • 法執行機関によるリアルタイム遠隔生体識別の利用に関する年次報告(第5条第7項) · Commission européenneCitoyenのカードはまだありません

更新情報を受け取る

この特集は、新しい数値を反映して、少なくとも四半期に1回更新されます。 フォローすると、更新のたびに通知を受け取れます。

出典

  1. [1]Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), OJ L, 2024/1689, 12.7.2024 –EUR-Lex, Official Journal of the European Union, 2024年7月12日
  2. [2]Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026 –EUR-Lex, Official Journal of the European Union, 2026年7月24日
  3. [3]Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), COM(2021) 206 final –European Commission (EUR-Lex), 2021年4月21日
  4. [4]Procedure 2021/0106/COD: steps of the procedure –EUR-Lex
  5. [5]Proposal for a Regulation amending Regulations (EU) 2024/1689 and (EU) 2018/1139 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI), COM(2025) 836 final –European Commission (EUR-Lex), 2025年11月19日
  6. [6]Procedure 2025/0359/COD: steps of the procedure –EUR-Lex
  7. [7]Artificial Intelligence Act: MEPs adopt landmark law (press release) –European Parliament, 2024年3月13日
  8. [8]AI Act: deal on simplification measures, ban on “nudifier” apps (press release) –European Parliament, 2026年5月7日
  9. [9]AI Act: EP approves simplification measures and “nudifier” app ban (press release) –European Parliament, 2026年6月16日
  10. [10]Legislative Train Schedule: Digital omnibus on AI, 2025/0359(COD) –European Parliament
  11. [11]The General-Purpose AI Code of Practice (contents and signatories) –European Commission, Shaping Europe’s digital future
  12. [12]The General-Purpose AI Code of Practice is now available –European Commission, Shaping Europe’s digital future
  13. [13]European AI Office –European Commission, Shaping Europe’s digital future
  14. [14]AI Act: regulatory framework for AI –European Commission, Shaping Europe’s digital future
  15. [15]Opinion of the European Economic and Social Committee on the Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) (2021/C 517/09) –European Economic and Social Committee (EUR-Lex, OJ C 517, 22.12.2021), 2021年12月22日
  16. [16]Opinion of the European Economic and Social Committee on the Digital Omnibus on AI and the Digital Omnibus (C/2026/3225) –European Economic and Social Committee (EUR-Lex, OJ C, 2.7.2026), 2026年7月2日
  17. [17]Opinion of the European Committee of the Regions: Digital simplification and Data Union Strategy (C/2026/3469) –European Committee of the Regions (EUR-Lex, OJ C, 24.7.2026), 2026年7月24日