« 世界経済フォーラム (WEF) 製造業におけるサイバーレジリエンス文化の構築 (2024.04.29) | Main | 世界経済フォーラム (WEF) 戦略的サイバーセキュリティ人材フレームワーク (2024.04.29) »

2024.05.03

米国 商務省 NISTがAIに関する4つのドラフトを公開し、意見募集を開始 (2023.04.29)

こんにちは、丸山満彦です。

AIに関する大統領令が発令されてから180日を迎える前に、お約束していた文書がNISTから発表され、意見募集が行われていますね...

文書は次の4つがあります...

生成的AIのリスク低減 NIST AI 600-1 Initial Public Draft Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile NIST AI 600-1 初期公開ドラフト 人工知能リスクマネジメントフレームワーク: 生成的人工知能プロファイル
AIシステムの訓練に使用されるデータへの脅威の軽減 NIST SP 800-218A (Initial Public Draft) Secure Software Development Practices for Generative AI and Dual-Use Foundation Models: An SSDF Community Profile NIST SP 800-218A(初期公開ドラフト) 生成的 AI とデュアルユース基盤モデルのための安全なソフトウェア開発プラクティス: SSDFコミュニティプロファイル
合成コンテンツのリスク低減 NIST AI 100-4 DRAFT FOR PUBLIC COMMENT Reducing Risks Posed by Synthetic Content An Overview of Technical Approaches to Digital Content Transparency NIST AI 100-4 パブリックコメント用ドラフト 合成コンテンツがもたらすリスクの低減 デジタルコンテンツの透明性に関する技術的アプローチの概要
AI標準に関するグローバルな取り組み NIST AI 100-5 Draft for public comment A Plan for Global Engagement on AI Standards NIST AI 100-5 パブリックコメント用ドラフト AI標準に関するグローバルな関与のための計画

 

商務省の発表...

U.S. Department of Commerce

・2024.04.29 Department of Commerce Announces New Actions to Implement President Biden’s Executive Order on AI

Department of Commerce Announces New Actions to Implement President Biden’s Executive Order on AI 商務省、AIに関するバイデン大統領の大統領令を実施するための新たな措置を発表
Announcements include draft guidance documents, a draft plan for international standards, and a new measurement program opening for public comment ガイダンス文書ドラフト、国際標準化計画ドラフト、新しい測定プログラムのパブリックコメント募集などが発表された。
The U.S. Department of Commerce announced today, following the 180-day mark since President Biden’s Executive Order (EO) on the Safe, Secure and Trustworthy Development of AI, several new announcements related to the EO. The Department’s National Institute of Standards and Technology (NIST) has released four draft publications intended to help improve the safety, security and trustworthiness of artificial intelligence (AI) systems. NIST has also launched a challenge series that will support development of methods to distinguish between content produced by humans and content produced by AI. In addition to NIST’s publications, Commerce’s U.S. Patent and Trademark Office (USPTO) is publishing a request for public comment (RFC) seeking feedback on how AI could affect evaluations of how the level of ordinary skills in the arts are made to determine if an invention is patentable under U.S. law, and earlier this year released guidance on the patentability of AI-assisted inventions. 米商務省は本日、バイデン大統領令「AIの安全・安心・信頼ある開発に関する大統領令(EO)」から180日を迎え、EOに関連するいくつかの新たな発表を行った。同省の国立標準技術研究所(NIST)は、人工知能(AI)システムの安全性、セキュリティ、信頼性を改善することを目的とした4つのドラフト機構を発表した。NISTはまた、人間が作成したコンテンツとAIが作成したコンテンツを区別する手法の開発を支援するチャレンジシリーズを開始した。NISTの出版物に加え、米国特許商標庁(USPTO)は、発明が米国法の下で特許可能かどうかを判断するために、技術分野における通常の技能の水準がどのように評価されるかにAIがどのような影響を及ぼす可能性があるかについての意見を求める意見要求(RFC)を公表しており、今年初めにはAI支援発明の特許性に関するガイダンスを発表した。
“In the six months since President Biden enacted his historic Executive Order on AI, the Commerce Department has been working hard to research and develop the guidance needed to safely harness the potential of AI, while minimizing the risks associated with it,” said U.S. Secretary of Commerce Gina Raimondo. “The announcements we are making today show our commitment to transparency and feedback from all stakeholders and the tremendous progress we have made in a short amount of time. With these resources and the previous work on AI from the Department, we are continuing to support responsible innovation in AI and America’s technological leadership.” 「バイデン大統領がAIに関する歴史的な大統領令を制定してから6ヶ月間、商務省は、AIに関連するリスクを最小限に抑えつつ、AIの可能性を安全に活用するために必要なガイダンスを研究・開発するために尽力してきた。「本日発表する内容は、すべての利害関係者からの透明性とフィードバックに対する我々のコミットメントを示すものであり、我々が短期間で多大な進歩を遂げたことを示している。これらのリソースとNISTのAIに関するこれまでの取り組みにより、我々はAIの責任あるイノベーションとアメリカの技術的リーダーシップを引き続き支援していく。
The NIST publications cover varied aspects of AI technology: The first two are guidance documents designed to help manage the risks of generative AI — the technology that enables chatbots and text-based image and video creation tools — and serve as companion resources to NIST’s AI Risk Management Framework (AI RMF) and Secure Software Development Framework (SSDF), respectively. A third NIST publication offers approaches for promoting transparency in digital content, which AI can alter; the fourth proposes a plan for developing global AI standards. These publications are initial drafts, which NIST is publishing now to solicit public feedback before submitting final versions later this year. NISTの出版物はAI技術の様々な側面をカバーしている: 最初の2つは、生成的AI(チャットボットやテキストベースの画像・動画作成ツールを実現する技術)のリスクマネジメントを支援するためのガイダンス文書で、それぞれNISTのAIリスクマネジメントフレームワーク(AI RMF)とセキュアソフトウェア開発フレームワーク(SSDF)の関連資料となっている。3つ目のNIST出版物は、AIが改変する可能性のあるデジタルコンテンツの透明性を促進するためのアプローチを提案している。これらの出版物は初期ドラフトであり、NISTは今年後半に最終版を提出する前に、一般からの意見を求めるために現在公表している。
“For all its potentially transformative benefits, generative AI also brings risks that are significantly different from those we see with traditional software,” said Under Secretary of Commerce for Standards and Technology and NIST Director Laurie E. Locascio. “These guidance documents will not only inform software creators about these unique risks, but also help them develop ways to mitigate the risks while supporting innovation.” 「標準技術担当商務次官兼NISTディレクターのローリー・E・ロカシオ氏は、「生成的AIは、潜在的に変革的な利益をもたらす一方で、従来のソフトウェアとは大きく異なるリスクももたらす。「これらのガイダンス文書は、このようなユニークなリスクについてソフトウェア作成者に知らせるだけでなく、イノベーションを支援しながらリスクを軽減する方法を開発するのにも役立つ。
USPTO is publishing an RFC seeking feedback on how AI could affect evaluations they make as they determine whether an invention is patentable under U.S. law. For example, the use of AI poses questions as to what qualifies as prior art and the assessment of the level of skill of a person having ordinary skill in the art. USPTO expects the responses to the RFC will help them evaluate the need for further guidance on these matters, aid in the development of any such guidance, and help inform USPTO’s work in the courts and in providing technical advice to Congress. USPTOはRFCを発行し、発明が米国法の下で特許可能かどうかを判断する際に、AIがどのように評価に影響するかについてのフィードバックを求めている。例えば、AIの使用は、何が先行技術として適格か、当業者の技術レベルの評価について疑問を投げかける。USPTOは、RFCへの回答が、これらの事項に関する更なるガイダンスの必要性を評価し、そのようなガイダンスの策定を支援し、裁判所におけるUSPTOの活動や議会への技術的助言の提供に役立つことを期待している。
“As AI assumes a larger role in innovation, we must encourage the responsible and safe use of AI to solve local and world problems and to develop the jobs and industries of the future, while ensuring AI does not derail the critical role IP plays in incentivizing human ingenuity and investment,” said Under Secretary of Commerce for Intellectual Property and Director of the USPTO Kathi Vidal. “This work builds on our inventorship guidance, which carefully set forth when the USPTO will issue a patent for AI-assisted innovations, and our continuing policy work at the intersection of AI and all forms of IP.” 「AIが技術革新においてより大きな役割を担うようになるにつれ、我々は、地域や世界の問題を解決し、将来の雇用や産業を発展させるために、責任ある安全なAIの利用を奨励しなければならないが、同時に、AIが、人間の創意工夫や投資を奨励する上で知的財産が果たす重要な役割を阻害しないようにしなければならない。「この作業は、USPTOがAIによるイノベーションに対して特許を発行する時期を慎重に定めた発明者ガイダンスと、AIとあらゆる形態の知的財産が交差する場所での継続的な政策作業に基づくものである。
More information on the announcements being made today can be found below. All four of the NIST publications are initial public drafts, and NIST is soliciting comments from the public on each by June 2, 2024. Instructions for submitting comments can be found in the respective publications. 本日発表された内容の詳細は以下を参照のこと。NISTの4つの出版物はすべて初期の公開ドラフトであり、NISTはそれぞれについて2024年6月2日までに一般からのコメントを募集している。コメントの提出方法は、それぞれの出版物に記載されている。
Mitigating the Risks of Generative AI 生成的AIのリスク低減
The AI RMF Generative AI Profile (NIST AI 600-1) can help organizations identify unique risks posed by generative AI and proposes actions for generative AI risk management that best aligns with their goals and priorities. Developed over the past year and drawing on input from the NIST generative AI public working group of more than 2,500 members, the guidance document is intended to be a companion resource for users of NIST’s AI RMF. The AI Profile guidance document centers on a list of 13 risks and more than 400 actions that developers can take to manage them AI RMFの生成的AIプロファイル(NIST AI 600-1)は、組織が生成的AIがもたらす固有のリスクを特定し、その目標や優先事項に最も適した生成的AIのリスクマネジメントのための行動を提案するのに役立つ。このガイダンス文書は、2,500人以上のメンバーからなるNISTの生成的AIパブリックワーキンググループからのインプットを基に、過去1年間かけて作成されたもので、NISTのAI RMFのユーザー向けのリソースとなることを意図している。AIプロファイル・ガイダンス・ドキュメントは、13のリスクのリストと、それらを管理するために開発者が取ることのできる400以上のアクションを中心としている。
The 13 risks include issues such as easier access to information related to chemical, biological, radiological or nuclear weapons; a lowered barrier to entry for hacking, malware, phishing, and other cybersecurity attacks; and the production of hate speech and toxic, denigrating or stereotyping content. Following the detailed descriptions of these 13 risks is a matrix of the 400 actions that developers can take to mitigate the risks. 13のリスクには、化学兵器、生物兵器、放射線兵器、核兵器に関連する情報へのアクセスの容易化、ハッキング、マルウェア、フィッシング、その他のサイバーセキュリティ攻撃への参入障壁の低下、ヘイトスピーチや有害、中傷的、ステレオタイプなコンテンツの生成といった問題が含まれている。これら13のリスクの詳細な説明の後に、開発者がリスクを軽減するために取ることができる400の行動のマトリックスがある。
Reducing Threats to the Data Used to Train AI Systems AIシステムの訓練に使用されるデータへの脅威の軽減
The second publication, Secure Software Development Practices for Generative AI and Dual-Use Foundation Models (NIST Special Publication (SP) 800-218A), is designed to be used alongside the SSDF (SP 800-218). While the SSDF is broadly concerned with securing the software’s lines of code, this companion resource expands the SSDF to help address concerns around malicious training data adversely affecting generative AI systems
.
2番目の刊行物である「生成的AIおよびデュアルユース基盤モデルのための安全なソフトウェア開発プラクティス」(NIST特別刊行物(SP)800-218A)は、SSDF(SP800-218)と並行して使用されるように設計されている。SSDFは、ソフトウェアのコード行の安全性に広く関係しているが、この付随リソースは、悪意のあるトレーニングデータが生成的AIシステムに悪影響を与えるという懸念に対処するためにSSDFを拡張している。
In addition to covering aspects of the training and use of AI systems, the new companion resource offers guidance on dealing with the training data and data collection process, including a matrix that identifies potential risk factors and strategies to address them. Among other recommendations, the document suggests analyzing training data for signs of poisoning, bias, homogeneity and tampering. AIシステムの訓練と使用の側面をカバーすることに加え、新しいコンパニオンリソースは、潜在的なリスク要因とそれに対処するための戦略を特定するマトリックスを含む、訓練データとデータ収集プロセスへの対処に関するガイダンスを提供している。他の推奨事項の中でも、この文書では、データ・ポイズニング、バイアス、均質性、改ざんの兆候についてトレーニングデータを分析することを提案している。
Reducing Synthetic Content Risks 合成コンテンツのリスク低減
Accompanying generative AI’s development has been the rise of “synthetic” content, which has been created or altered by AI. Offering technical approaches for promoting transparency in digital content is the goal of NIST’s new draft publication, Reducing Risks Posed by Synthetic Content (NIST AI 100-4). This publication informs, and is complementary to, a separate report on understanding the provenance and detection of synthetic content that AI EO Section 4.5(a) tasks NIST with providing to the White House. 生成的AIの発展に伴い、AIによって作成または改変された「合成」コンテンツが増加している。デジタルコンテンツの透明性を促進するための技術的アプローチを提供することが、NISTの新しいドラフト出版物「合成コンテンツがもたらすリスクの低減」(NIST AI 100-4)の目的である。この出版物は、AI EO 4.5(a)項がNISTにホワイトハウスへのプロバイダを命じている、合成コンテンツの出所と検知の理解に関する別の報告書に情報を提供し、それを補完するものである。
NIST AI 100-4 lays out methods for detecting, authenticating and labeling synthetic content, including digital watermarking and metadata recording, where information indicating the origin or history of content such as an image or sound recording is embedded in the content to assist in verifying its authenticity. The report does not focus only on the dangers of synthetic content; it is intended to reduce risks from synthetic content by understanding and applying technical approaches for improving the content’s transparency, based on use case and context. NIST AI 100-4 は、電子透かしやメタデータの記録(画像や録音などのコンテンツの出所や履歴を示す情報をコンテンツに埋め込んで、その真正性を検証するのに役立てる)など、合成コンテンツの検出、本人認証、ラベリングの方法を示している。この報告書は、合成コンテンツの危険性だけに焦点を当てるのではなく、ユースケースやコンテクストに基づき、コンテンツの透明性を改善するための技術的アプローチを理解し、適用することで、合成コンテンツによるリスクを低減することを意図している。
Global Engagement on AI Standards AI標準に関するグローバルな取り組み
AI systems are transforming American society and around the world. A Plan for Global Engagement on AI Standards (NIST AI 100-5) is designed to drive the worldwide development and implementation of AI-related consensus standards, cooperation and coordination, and information sharing. AIシステムはアメリカ社会と世界を変革しつつある。AI標準に関するグローバルな関与のための計画(NIST AI 100-5)は、AI関連のコンセンサス標準の世界的な開発・実施、協力・調整、情報共有を推進するためのものである。
The draft invites feedback on areas and topics that may be urgent for AI standardization. Such topics, which are ready for standardization, include mechanisms for enhancing awareness of the origin of digital content, whether authentic or synthetic; and shared practices for testing, evaluation, verification and validation of AI systems. Other topics may require more scientific research and development to establish a foundational scientific understanding about critical components of the potential standard. このドラフトでは、AIの標準化が急務と思われる分野やトピックについて意見を募集している。標準化の準備が整ったこのようなトピックには、本物か合成かを問わず、デジタルコンテンツの出所に関する認識を高める仕組みや、AIシステムのテスト、評価、検証、妥当性確認のための共有プラクティスなどが含まれる。その他のトピックについては、潜在的な標準の重要な構成要素に関する基礎的な科学的理解を確立するために、より科学的な研究開発が必要となる可能性がある。
NIST GenAI NIST GenAI
In addition to the four documents, NIST is also announcing NIST GenAI, a new program to evaluate and measure generative AI technologies. The program is part of NIST's response to the Executive Order, and its efforts will help inform the work of the U.S. AI Safety Institute at NIST. 4つの文書に加え、NISTは生成的AI技術を評価・測定する新しいプログラム、NIST GenAIも発表している。このプログラムは大統領令に対するNISTの対応の一環であり、その機構はNISTの米国AI安全研究所の作業に情報を提供するのに役立つ。
The NIST GenAI program will issue a series of challenge problems designed to evaluate and measure the capabilities and limitations of generative AI technologies. These evaluations will be used to identify strategies to promote information integrity and guide the safe and responsible use of digital content. One of the program’s goals is to help people determine whether a human or an AI produced a given text, image, video or audio recording. Registration opens in May for participation in the pilot evaluation, which will seek to understand how human-produced content differs from synthetic content. More information about the challenge and how to register can be found on the NIST GenAI website. NISTのGenAIプログラムは、生成的AI技術の能力と限界を評価・測定するために設計された一連のチャレンジ問題を発行する。これらの評価は、情報の完全性を促進し、デジタルコンテンツの安全で責任ある利用を導くための戦略を特定するために使用される。このプログラムの目標のひとつは、与えられたテキスト、画像、ビデオ、オーディオ録音が、人間によって作られたのか、AIによって作られたのかを人々が判断できるようにすることである。パイロット評価への参加登録は5月に開始され、人間が制作したコンテンツと合成コンテンツとの違いを理解することを目指す。チャレンジの詳細と登録方法は、NIST GenAIのウェブサイトを参照のこと。
USPTO RFC on Patentability 特許性に関するUSPTO RFC
The full Federal Register Notice can be found HERE. Comments are due July 29, 2024. Please see the Federal Register Notice for instructions on submitting comments. 連邦官報告示の全文はこちらを参照のこと。意見の提出期限は2024年7月29日である。意見の提出方法については、連邦官報告示を参照のこと。
To incentivize, protect, and encourage investment in innovations made possible through the use of artificial intelligence (AI), and to provide the clarity to the public on the patentability of AI-assisted inventions, the USPTO has previously published guidance in the Federal Register. 人工知能(AI)を利用したイノベーションへの投資を奨励し、保護し、奨励するため、また、AIを利用した発明の特許性を明確にするため、USPTOはこれまで連邦官報でガイダンスを発表してきた。

 

 

NISTの発表...

NIST - ITL

・2024.04.29 Department of Commerce Announces New Actions to Implement President Biden’s Executive Order on AI

 

Department of Commerce Announces New Actions to Implement President Biden’s Executive Order on AI 商務省、AIに関するバイデン大統領の大統領令を実施するための新たな措置を発表
Announcements include draft guidance documents, a draft plan for international standards, and a new measurement program opening for public comment. ガイダンス文書ドラフト、国際標準化計画ドラフト、新しい測定プログラムのパブリックコメント募集などが発表された。
The U.S. Department of Commerce announced today, following the 180-day mark since President Biden’s Executive Order (EO) on the Safe, Secure and Trustworthy Development of AI, several new announcements related to the EO. The department’s National Institute of Standards and Technology (NIST) has released four draft publications intended to help improve the safety, security and trustworthiness of artificial intelligence (AI) systems. NIST has also launched a challenge series that will support development of methods to distinguish between content produced by humans and content produced by AI. In addition to NIST’s publications, Commerce’s U.S. Patent and Trademark Office (USPTO) is publishing a request for public comment (RFC) seeking feedback on how AI could affect evaluations of how the level of ordinary skills in the arts are made to determine if an invention is patentable under U.S. law, and earlier this year released guidance on the patentability of AI-assisted inventions. 米国商務省は本日、バイデン大統領令「AIの安全・安心・信頼ある開発に関する大統領令(EO)」から180日を経過したことを受け、EOに関連するいくつかの新たな発表を行った。同省の国立標準技術研究所(NIST)は、人工知能(AI)システムの安全性、セキュリティ、信頼性を改善することを目的とした4つのドラフト機構を発表した。NISTはまた、人間が作成したコンテンツとAIが作成したコンテンツを区別する手法の開発を支援するチャレンジシリーズを開始した。NISTの発表に加え、商務省の米国特許商標庁(USPTO)は、発明が米国法の下で特許可能かどうかを判断するために、技術分野における通常の技能の水準がどのように評価されるかにAIがどのような影響を及ぼす可能性があるかについての意見を求める意見要求(RFC)を発表しており、今年初めにはAI支援発明の特許性に関するガイダンスを発表した。
“In the six months since President Biden enacted his historic Executive Order on AI, the Commerce Department has been working hard to research and develop the guidance needed to safely harness the potential of AI, while minimizing the risks associated with it,” said U.S. Secretary of Commerce Gina Raimondo. “The announcements we are making today show our commitment to transparency and feedback from all stakeholders and the tremendous progress we have made in a short amount of time. With these resources and the previous work on AI from the department, we are continuing to support responsible innovation in AI and America’s technological leadership.” 「バイデン大統領がAIに関する歴史的な大統領令を制定してから6ヶ月間、商務省は、AIに関連するリスクを最小限に抑えつつ、AIの可能性を安全に活用するために必要なガイダンスの研究と開発に尽力してきた。「本日発表する内容は、すべての利害関係者からの透明性とフィードバックに対する我々のコミットメントを示すものであり、我々が短期間で多大な進歩を遂げたことを示している。これらのリソースと、同省のAIに関するこれまでの取り組みにより、我々はAIの責任あるイノベーションとアメリカの技術的リーダーシップを支援し続けている。"
The NIST publications cover varied aspects of AI technology: The first two are guidance documents designed to help manage the risks of generative AI — the technology that enables chatbots and text-based image and video creation tools — and serve as companion resources to NIST’s AI Risk Management Framework (AI RMF) and Secure Software Development Framework (SSDF), respectively. A third NIST publication offers approaches for promoting transparency in digital content, which AI can alter; the fourth proposes a plan for developing global AI standards. These publications are initial drafts, which NIST is publishing now to solicit public feedback before submitting final versions later this year. NISTの出版物はAI技術の様々な側面をカバーしている: 最初の2つは、生成的AI(チャットボットやテキストベースの画像・動画作成ツールを可能にする技術)のリスクマネジメントを支援するためのガイダンス文書で、それぞれNISTのAIリスクマネジメントフレームワーク(AI RMF)とセキュアソフトウェア開発フレームワーク(SSDF)の関連資料となっている。3つ目のNIST出版物は、AIが改変する可能性のあるデジタルコンテンツの透明性を促進するためのアプローチを提案している。これらの出版物は初期ドラフトであり、NISTは今年後半に最終版を提出する前に、一般からの意見を募るために現在公表している。
“For all its potentially transformative benefits, generative AI also brings risks that are significantly different from those we see with traditional software. These guidance documents will not only inform software creators about these unique risks, but also help them develop ways to mitigate the risks while supporting innovation.” —Under Secretary of Commerce for Standards and Technology and NIST Director Laurie E. Locascio 「生成的AIは、その潜在的な利点のすべてにおいて、従来のソフトウェアとは大きく異なるリスクももたらす。これらのガイダンス文書は、ソフトウェア作成者にこれらのユニークなリスクについて知らせるだけでなく、イノベーションを支援しながらリスクを軽減する方法を開発する助けにもなる。」-標準技術担当商務次官兼NIST理事 ローリー・E・ロカシオ

 


 

1. 生成的AIのリスク低減

・2024.04.29 [PDF] NIST AI 600-1 Initial Public Draft Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

20240502-221322

 

NIST AI 600-1 Initial Public Draft Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile NIST AI 600-1 初期公開ドラフト 人工知能リスクマネジメントフレームワーク: 生成的人工知能プロファイル
1. Introduction. 1. 序文
1.1. About this Profile 1.1. このプロファイルについて
2. Overview of Risks Unique to or Exacerbated by GAI 2. GAI特有の、あるいはGAIによって悪化するリスクの概要
◦2.1. CBRN Information ◦2.1. CBRN情報
◦2.2. Confabulation ◦2.2. 混同
◦2.3. Dangerous or Violent Recommendations. ◦2.3. 危険または暴力的な勧告。
◦2.4. Data Privacy ◦2.4. データ・プライバシー
◦2.5. Environmental ◦2.5. 環境 ◦2.6.
◦2.6. Human-Al Configuration ◦2.6. 人間-アルコンフィギュレーション
◦2.7. Information Integrity ◦2.7. 情報の完全性
◦2.8. Information Security ◦2.8. 情報セキュリティ
◦2.9. Intellectual Property ◦2.9. 知的財産
◦2.10. Obscene, Degrading, and/or Abusive Content ◦2.10. わいせつ、品位低下、および/または虐待的なコンテンツ
◦2.11. Toxicity, Bias, and Homogenization ◦2.11. 毒性、バイアス、均質化 ◦2.12.
◦2.12. Value Chain and Component Integration ◦2.12. バリューチェーンとコンポーネントの統合
3. Actions to Manage GAI Risks 3. GAIリスクをマネジメントするための行動
Appendix A. Primary GAI Considerations  附属書A. GAIに関する主な考慮事項 
Appendix B. References 附属書B. 参考文献

 

 


2. AIシステムの訓練に使用されるデータへの脅威の軽減

・2024.04.29 NIST SP 800-218A (Initial Public Draft) Secure Software Development Practices for Generative AI and Dual-Use Foundation Models: An SSDF Community Profile

NIST SP 800-218A (Initial Public Draft) Secure Software Development Practices for Generative AI and Dual-Use Foundation Models: An SSDF Community Profile NIST SP 800-218A(初期公開ドラフト) 生成的 AI とデュアルユース基盤モデルのための安全なソフトウェア開発プラクティス: SSDFコミュニティプロファイル
Announcement 発表
This publication augments the secure software development practices and tasks defined in SP 800-218, Secure Software Development Framework (SSDF) Version 1.1: Recommendations for Mitigating the Risk of Software Vulnerabilities. SP 800-218A adds practices, tasks, recommendations, considerations, notes, and informative references that are specific to AI model development throughout the software development life cycle.  本書は、SP 800-218「セキュアソフトウェア開発フレームワーク(SSDF)バージョン 1.1」で定義されたセキュアソフトウェア開発の実践とタスクを補強するものである: ソフトウェアの脆弱性リスクを低減するための推奨事項である。SP 800-218A は、ソフトウェア開発ライフサイクル全体を通して、AI モデル開発に特化した実践、タスク、推奨、考慮事項、注記、および参考文献を追加している。
These additions are documented in the form of an SSDF Community Profile to support Executive Order (EO) 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, which tasked NIST with “developing a companion resource to the [SSDF] to incorporate secure development practices for generative AI and for dual-use foundation models.”  これらの追加事項は、大統領令(EO)14110「人工知能の安全、確実、信頼できる開発と使用」をサポートするために、SSDFコミュニティ・プロファイルの形で文書化されている。このコミュニティ・プロファイルは、NISTに「生成的AIとデュアルユース基盤モデルのための安全な開発プラクティスを組み込むために、[SSDF]に付随するリソースを開発する」ことを課している。
This Community Profile is intended to be useful to the producers of AI models, the producers of AI systems that use those models, and the acquirers of those AI systems. This Profile should be used in conjunction with SP 800-218. このコミュニティ・プロファイルは、AI モデルの製作者、それらのモデルを使用する AI システムの製作者、およびそれらの AI システムの取得者にとって有用であることを意図している。このプロファイルは SP 800-218 と併せて使用されるべきである。
Abstract 概要
This document augments the secure software development practices and tasks defined in Secure Software Development Framework (SSDF) version 1.1 by adding practices, tasks, recommendations, considerations, notes, and informative references that are specific to AI model development throughout the software development life cycle. These additions are documented in the form of an SSDF Community Profile to support Executive Order (EO) 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, which tasked NIST with “developing a companion resource to the [SSDF] to incorporate secure development practices for generative AI and for dual-use foundation models.” This Community Profile is intended to be useful to the producers of AI models, the producers of AI systems that use those models, and the acquirers of those AI systems. This Profile should be used in conjunction with NIST Special Publication (SP) 800-218, Secure Software Development Framework (SSDF) Version 1.1: Recommendations for Mitigating the Risk of Software Vulnerabilities. 本文書は、セキュアソフトウェア開発フレームワーク(SSDF)バージョン1.1で定義されたセキュアソフトウェア開発プラクティスとタスクを補強するものであり、ソフトウェア開発ライフサイクル全体を通してAIモデル開発に特化したプラクティス、タスク、推奨事項、考慮事項、注釈、参考文献を追加するものである。これらの追加事項は、大統領令(EO)14110「人工知能の安全、セキュア、かつ信頼できる開発と使用」をサポートするために、SSDFコミュニティ・プロファイルの形で文書化されている。このコミュニティ・プロファイルは、NISTに「生成的AIとデュアルユース基盤モデルのための安全な開発手法を取り入れた(SSDFの)付属リソースを開発する」ことを課している。このコミュニティ・プロファイルは、AI モデルの製作者、それらのモデルを使用する AI システムの製作者、およびそれらの AI システムの取得者にとって有用であることを意図している。このプロファイルは、NIST 特別刊行物(SP)800-218「安全なソフトウェア開発フレームワーク(SSDF)バージョン 1.1」と併せて使用されるべきである: ソフトウェアの脆弱性リスクを低減するための推奨事項" と併せて使用すること。

 

・[PDF] NIST.SP.800-218A.ipd

20240502-221733

 

Note to Reviewers  査読者への注記 
NIST welcomes feedback on any part of this document but is particularly interested in the following: NIST はこの文書のどの部分についてのフィードバックも歓迎するが、特に次の点に関心をもっている:
• The Profile suggests adding several practices and tasks to those defined in SSDF version 1.1. Are these additions reasonable? What other additions would help address secure development practices for generative AI and dual-use foundation models? - このプロファイルは、SSDF version 1.1 で定義されたものに、いくつかの対策とタスクを追加することを提案している。これらの追加は妥当か?生成的 AI とデュアルユース基盤モデルのための安全な開発手法に対処するために、他にどのような追加事項が役立つか?
• What changes should be made to the Profile’s Recommendations, Considerations, and Notes column to help address secure software development practices for generative AI and dual-use foundation models? ・生成的 AI とデュアルユース基盤モデルのための安全なソフトウェア開発手法に対処するために,プロファイルの「推奨事項」,「考慮事項」,「注意事項」欄にどのような変更を加えるべきか?
• What additional cybersecurity, privacy, and/or reproducibility considerations should be taken into account when selecting model training techniques (e.g., deterministic model training)? ・モデル・トレーニング技術(例えば、決定論的モデル・トレーニング)を選択する際に、サイバーセキュリティ、プライバシー、および/または再現性について、どのような追加的な考慮が必要か?
• What suggestions do you have for Implementation Examples and additional Informative References for the Profile? ・このプロファイルの実装例や追加的な参考文献についてどのような提案があるか?
• Is this Profile fully applicable to the secure development of other types of AI models besides generative and dual-use foundation models? If not, what changes could be made to the Profile to accommodate other AI models? ・このプロファイルは生成的AIモデルや両用基礎モデル以外の他のタイプのAIモデルの安全な開発にも完全に適用できますか?もしそうでないなら,他の AI モデルに対応するためにこのプロファイルにどのような変更を加えることができるか。
• Is this Profile flexible enough to support AI model producers, the producers of AI systems using those models, and the acquirers of those AI systems? ・このプロファイルは,AIモデルの製作者,それらのモデルを使用するAIシステムの製作者,それらのAIシステムの取得者を支援するのに十分な柔軟性があるか。
• What guidance, templates, or other resources on SSDF Community Profile use would you find beneficial?  ・ SSDF コミュニティ・プロファイルの使用に関して,どのようなガイダンス,テンプレート,またはその他のリソースが有益であると思われるか?
If you are from a standards developing organization (SDO) or another organization that is defining a set of secure practices for AI model development and you would like to map your standard or guidance to the SSDF profile, please contact the authors at ssdf@nist.gov. They will introduce you to the National Online Informative References Program (OLIR), where you can submit your mapping to augment the existing set of informative references. もしあなたが標準開発組織(SDO)や、AIモデル開発のための一連の安全な実践方法を定義している他の組織の方で、その標準やガイダンスをSSDFプロファイルに対応させたいとお考えであれば、認可担当者(ssdf@nist.gov)までご連絡いただきたい。このプログラムでは、既存の参考文献を補強するためのマッピングを提出することができる。

 

目次...

1. Introduction. 1. 序文
1.1. Purpose 1.1. 目的
1.2. Scope 1.2. 適用範囲
1.3. Sources of Expertise 1.3. 専門家の情報源
1.4. Document Structure 1.4. 文書の構成
Using the SSDF Community Profile SSDF コミュニティプロファイルの使用
SSDF Community Profile for Al Model Development Al モデル開発のための SSDF コミュニティプロファイル
References 参考文献
Appendix A. Glossary 附属書 A. 用語集
List of Tables 表のリスト
Table 1. SSDF Community Profile for Al Model Development 表1. Al モデル開発のための SSDF コミュニティプロファイル

 

 


 

3. 合成コンテンツのリスク低減

・[PDF] NIST AI 100-4 DRAFT FOR PUBLIC COMMENT Reducing Risks Posed by Synthetic Content An Overview of Technical Approaches to Digital Content Transparency

20240502-221427

 

目次...

NIST AI 100-4 DRAFT FOR PUBLIC COMMENT Reducing Risks Posed by Synthetic Content An Overview of Technical Approaches to Digital Content Transparency NIST AI 100-4 パブリックコメント用ドラフト 合成コンテンツがもたらすリスクの低減 デジタルコンテンツの透明性に関する技術的アプローチの概要
1. Summary 1. 概要
2. Harms and Risks from Synthetic Content 2. 合成コンテンツによる危害とリスク
3. Current Approaches, Issues, and Opportunities 3. 現在のアプローチ、課題、機会
3.1. Provenance Data Tracking 3.1. 証明データの追跡
3.1.1. Digital Watermarking 3.1.1. 電子透かし
3.1.2. Metadata Recording 3.1.2. メタデータの記録
3.1.3. Effectiveness of Provenance Data Tracking Techniques Across Different Types of Content 3.1.3. 様々な種類のコンテンツにおける証明データ追跡技術の有効性
3.1.4. Synthetic Content Detection 3.1.4. 合成コンテンツの検知
4. Testing and Evaluating Provenance Data Tracking and Synthetic Content Detection Techniques 4. プロバンスデータ追跡と合成コンテンツ検知技術のテストと評価
4.1. Testing and Evaluating Provenance Data Tracking Techniques 4.1. 実証データ追跡技術のテストと評価
4.1.1. Testing and Evaluating Digital Watermarking Techniques 4.1.1. 電子透かし技術のテストと評価
4.1.2. Testing and Evaluating Metadata Recording Techniques 4.1.2. メタデータ記録技術のテストと評価
4.2. Testing and Evaluating Synthetic Content Detection Techniques 4.2. 合成コンテンツ検知技術のテストと評価
4.2.1. Testing and Evaluating Automated Content-Based Detection Techniques  4.2.1. 自動コンテンツ検知技術のテストと評価 
4.2.2. Testing and Evaluating Provenance Detection Techniques  4.2.2. プロベナンス検知技術のテストと評価 
4.2.3. Testing and Evaluating Human-Assisted Detection Techniques 4.2.3. 人間支援検知技術のテストと評価
4.3. Additional Issues for Consideration  4.3. その他の検討課題 
5. Preventing and Reducing Harms from Synthetic Child Sexual Abuse Material and Non-Consensual Intimate Imagery 5. 合成児童性的虐待素材および非合意的親密画像による被害の防止と軽減
5.1. Current Technical Mitigations to Prevent and Reduce Harms from Synthetic CSAM and NCIl 5.1. 合成 CSAM および NCIl による危害を防止・低減するための現在の技術的低減策
5.1.1. Training Data Filtering 5.1.1. 訓練データのフィルタリング
5.1.2. Input Data Filtering 5.1.2. 入力データのフィルタリング
5.1.3. Image Output Filtering 5.1.3. 画像出力フィルタリング
5.1.4. Hashing Confirmed Synthetic CSAM and NCII 5.1.4. 合成CSAMとNCIIのハッシュ化確認
5.1.5. Provenance Data Tracking Techniques for Synthetic CSAM and NCII 5.1.5. 合成CSAMとNCIIの証明データ追跡技術
5.1.6. Red-Teaming and Testing for CSAM and NCII 5.1.6. CSAMとNCIIのレッドチームとテスト
6. Application of Concepts to the NIST Al Risk Management Framework Lifecycle 6. NIST Alリスクマネジメントフレームワークライフサイクルへの概念の適用
7. Conclusion  7. 結論 
8. Bibliography 8. 参考文献
Appendix A. Current Standards 附属書 A. 現行標準
Appendix B. Technical Tools 附属書B. テクニカルツール
Appendix C. Provenance Data Tracking 附属書C. 証拠データの追跡
Appendix D. Synthetic Content Detection 附属書D. 合成コンテンツの検知
Appendix E. Testing and Evaluation 附属書E. テストと評価
Appendix F. Glossary 附属書F. 用語集

 

 


 

4. AI標準に関するグローバルな取り組み

・2024.04.29 [PDF] NIST AI 100-5 Draft for public comment A Plan for Global Engagement on AI Standards

20240502-224339

 

NIST AI 100-5 Draft for public comment A Plan for Global Engagement on AI Standards NIST AI 100-5 パブリックコメント用ドラフト AI標準に関するグローバルな関与のための計画
1. Executive Summary 1. エグゼクティブサマリー
2. Introduction 2. 序文
3. Objectives for Engagement on Al Standards 3. 標準に関するエンゲージメントの目的
3.1. Scientifically sound Al standards that are accessible and amenable to adoption 3.1. 科学的に健全で、アクセスしやすく、採用しやすいAl標準。
3.2. Al standards that reflect the needs and inputs of diverse global stakeholders 3.2. 世界の多様な利害関係者のニーズと意見を反映した標準。
3.3. Al standards that are developed in a process that is open, transparent, and driven by consensus 3.3. オープンで透明性が高く、コンセンサスによって推進されるプロセスで開発される標準。
3.4. International relationships that are strengthened by engagement on Al standards 3.4. 標準への関与によって強化される国際関係
4. Priority Topics for Standardization Work 4. 標準化作業の優先課題
4.1. Urgently needed and ready for standardization  4.1. 緊急に必要とされ、標準化の準備が整っているもの 
4.2. Needed, but requiring more scientific work before standardization 4.2. 標準化は必要だが、標準化の前にさらに科学的な作業が必要である。
4.3. Needed, but requiring significant foundational work 4.3. 必要であるが、重要な基礎作業が必要である
5. Recommended Global Engagement Activities 5. 推奨されるグローバル関与活動
5.1. Prioritize engagement in SDOs, including research and related technical activities 5.1. 研究および関連する技術活動を含め、SDOへの参加を優先する。
5.2. Facilitate diverse multistakeholder engagement in Al standards development 5.2. Al 規格開発における多様なマルチステークホルダーの関与を促進する。
5.3. Promote global alignment on Al standards approaches 5.3. Al 規格のアプローチに関するグローバルな連携を促進する。
Appendix A. Standards in Relation to Al 附属書 A. Alに関連する標準
A.1. What are standards and why are they important? A.1. 標準とは何か、なぜ重要なのか?
A.2. How are standards developed? A.2. 標準はどのように開発されるのか?
A.3. How do Al standards differ from other technical standards? A.3. Alの標準は、他の技術標準とどのように違うのか?
Appendix B. The Current Landscape of Al Standardization 附属書B. Al標準化の現状
B.1. Horizontal standards: SDOs and topics B.1. 水平標準: SDO とトピック
B.2. Participation in Al standards development B.2. Al 規格開発への参加

 


 

こちらも...

Federal Register 

・2024.04.30 Notice Request for Comments Regarding the Impact of the Proliferation of Artificial Intelligence on Prior Art, the Knowledge of a Person Having Ordinary Skill in the Art, and Determinations of Patentability Made in View of the Foregoing

人工知能の普及が先行技術、当該技術分野における通常の知識を有する者の知識、およびこれらを考慮した特許性の判断に与える影響に関する意見要求...

その際には、参考に...

AI支援発明の発明者ガイダンス

・2024.02.13 Inventorship Guidance for AI-Assisted Inventions

 


 

まるちゃんの情報セキュリティ気まぐれ日記

・2024.04.01 米国 財務省 金融サービスセクターにおける人工知能特有のサイバーセキュリティ・リスクの管理

・2024.03.31 米国 連邦政府機関の人工知能利用におけるガバナンス、イノベーション、リスク管理を推進するためのOMBの方針

・2024.03.28 米国 国防総省 国防高等研究計画局(DARPA)の重要な目標は、国防総省にとって信頼できる人工知能を開発することである。

・2024.03.02 米国 連邦金融機関審査委員会(FFIEC) 住宅ローンにおける評価の差別とバイアスに関連する審査原則に関する声明 (2024.02.12)

・2024.02.17 米国 NIST 「非倫理的な被験者研究における過去の過ちを避ける: 人工知能の原理から実践への移行」

・2024.02.13 米国 AI安全研究所を設立

・2024.01.31 米国 ホワイトハウス 主要なAI行動を発表

・2024.01.28 AI使用に関する国際ガイダンス

・2024.01.28 米国 ニューヨーク州 金融サービス局 意見募集 保険引受と価格設定における人工知能システムと外部消費者データおよび情報源の利用に関する通達案 (2024.01.17)

・2023.12.15 連邦政府テクノロジー・リーダーがOMBのAI政策ドラフトについて知っておくべきことトップ10

・2023.10.31 米国 人工知能の安全、安心、信頼できる開発と利用に関する大統領令

 

 

 

 

 

|

« 世界経済フォーラム (WEF) 製造業におけるサイバーレジリエンス文化の構築 (2024.04.29) | Main | 世界経済フォーラム (WEF) 戦略的サイバーセキュリティ人材フレームワーク (2024.04.29) »

Comments

Post a comment



(Not displayed with comment.)


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



« 世界経済フォーラム (WEF) 製造業におけるサイバーレジリエンス文化の構築 (2024.04.29) | Main | 世界経済フォーラム (WEF) 戦略的サイバーセキュリティ人材フレームワーク (2024.04.29) »