국방 AI의 보안성 강화를 위한 보안정책 및 연구개발 전략 수립방향
Security Policy and R&D Directions to Enhance Security for Defense AI
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
AI 기술의 윤리와 책임에 대한 국제적 논의가 활발하게 이루어짐에 따라서 AI에 대한 보안 강화를 위한 실질적인 방안을 정립할 필요성이 제기되고 있다. 특히 AI 기술이 탑재된 국방 시스템은 국가 안보와 국민 생활에 미치는 잠재적 영향력이 크기 때문에 다른 분야보다 높은 보안이 요구된다. 본 논문에서는 국방 AI의 개발과 활용 과정에서 발생할 수 있는 보안 취약점을 예방하고 보안위험을 관리하기 위한 정책 방향을 제시하고, 이러한 정책을 추진하는데 필요한 연구개발 전략의 수립방향을 제시한다. 이를 위해 국내외에서 개발된 AI 보안에 관한 가이드에 제시된 보안대책을 분석하여 국방 보안정책에 반영해야 할 요소를 식별하고 국방 보안제도의 특수성을 고려하여 별도로 정립해야할 특화 보안기준을 제시한다. 또한, 높은 보안성을 요구하는 국방의 특수성을 고려하여 구현된 보안기준의 적절성을 검증하고 AI 고유의 취약점을 효율적으로 관리함으로써 국방 AI의 보안 강화를 위한 연구개발 과제의 전략적 방향을 제시한다.
Abstract
As international discussions on the ethics and responsibility of AI technology are actively taking place, practical measures must be developed to strengthen the security of AI. In particular, defense systems equipped with AI technology have a high potential impact on national security and people's lives, so higher security must be required than in other fields. This paper presents policy directions for preventing security vulnerabilities that may arise during the development and use of defense AI and managing security risks, as well as directions for establishing research and development strategies necessary for policy implementation. To this end, we analyze the security measures presented in guides on the security of AI developed domestically and overseas. Also, we propose specific security requirements for defense domain considering characteristics of the defense security policy. Moreover, we present the strategic direction of research and development tasks necessary to strengthen security and institutionalize future defense AI security by verifying the appropriateness of implemented security measures and managing vulnerabilities of AI efficiently.
참고문헌 (27)
- M. Fredrikson, "Model inversion attacks that exploit confidence information and basic countermeasures", Proceedings of the 22nd ACM SIGSAC conference on computer and communications security, pp.1322~1333 (2015)
- "Military AI ignited by the Ukraine war is a 'double-edged sword'...in the international community's debate", Yonhap News, ~ (2023)
- "U.S. National Intelligence Service 'AI Security Center'... seems to be aware of hacking", The Chosun Daily, ~ (2023)
- The Dong-A Ilbo, "Cooperation to prevent catastrophic hazards of AI... First joint declaration by 28 countries, including the U.S., U,K., and China", ~ (2023)
- Cybersecurity and Infrastructure Security Agency (CISA), "2023-2024CISA roadmap for artificial intelligence", ~ (2023)
- National Cyber Security Centre (NCSC), "Guidelines for secure AI system development", ~ (2023)
- Operational feasibility of adversarial attacks against artificial intelligence, "L.A. Zhang", RAND Corporation, ~, pp.RR-A866-1 (2022)
- International Organization for Standardization and International Electrotechnical Commission, "Overview of trustworthiness in artificial intelligence", ISO, 2020
- Securing Artificial Intelligence; Mitigation Strategy Report, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104 222 V1.2.1 (2024)
- Securing Artificial Intelligence; Security Testing of AI, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104 066 V1.1.1 (2024)
- Securing Artificial Intelligence; Automated Manipulation of Multimedia Identity Representations, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104 062 V1.2.1 (2024)
- Securing Artificial Intelligence TC (SAI); Privacy aspects of AI, "ML systems", European Telecommunications Standards Institute, 2024, ETSI TR 104 225, V1.1.1
- Securing Artificial Intelligence (SAI); Proofs of Concepts Framework, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104 067 V1.1.1 (2024)
- Securing Artificial Intelligence (SAI); Traceability of AI Models, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104 032 V1.1.1 (2022)
- Securing Artificial Intelligence (SAI); Collaborative Artificial Intelligence, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI TR 104031 V1.1.1 (2024)
- Securing Artificial Intelligence (SAI); Data Supply Chain Security, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI GR SAI 002 V1.1.1 (2021)
- Securing Artificial Intelligence (SAI); Explicability and transparency of AI processing, "European Telecommunications Standards Institute", ETSI, ~, pp.ETSI GR SAI 007 V1.1.1 (2023)
- AI risk management framework, "National Institute of Standards and Technology (NIST)", NIST, ~, pp.NIST AI 100-1 (2023)
- Adversarial machine learning-a taxonomy and terminology of attacks and mitigations, "National Institute of Standards and Technology (NIST)", NIST, ~, pp.NIST AI 100-2 (2023)
- ALTAI -the assessment list on trustworthy artificial intelligence, "European Commission (EC)", ~, pp.KK-02-20-479-EN-N (2020)
- Securing machine learning algorithms, "European Union Agency for Cybersecurity (ENISA)", ~, pp.TP-06-21-153-ENN (2021)
- Ministry of Science and ICT, "Guide for trustworthy artificial intelligence development" (2023), Jinhan M&B INC
- National Intelligence Service (NIS), "Information security considerations related to the introduction and operation of artificial intelligence systems in state and public institutions", ~ (2019)
- National Intelligence Service (NIS), "Security guidelines using generative artificial intelligence such as ChatGPT", ~ (2023)
- Security guidelines for the artificial intelligence in financial services, "Financial Security Institute (FSI)", ~, pp.AGR-VIII-2023-2-405 (2023)