[Paper Review] Adapting cybersecurity frameworks to manage frontier AI risks: A defense-in-depth approach

Shaun Ee, J.N. O'Brien+4 more|arXiv (Cornell University)|Aug 15, 2024

Information and Cyber Security4 citations

TL;DR

This paper proposes a defense-in-depth strategy for managing frontier AI risks by adapting established cybersecurity frameworks—specifically the NIST Cybersecurity Framework and AI Risk Management Framework—through three complementary approaches: functional, lifecycle, and threat-based. It recommends adopting the functional approach immediately, while developing detailed AI lifecycle models and threat intelligence databases for future resilience.

ABSTRACT

The complex and evolving threat landscape of frontier AI development requires a multi-layered approach to risk management ("defense-in-depth"). By reviewing cybersecurity and AI frameworks, we outline three approaches that can help identify gaps in the management of AI-related risks. First, a functional approach identifies essential categories of activities ("functions") that a risk management approach should cover, as in the NIST Cybersecurity Framework (CSF) and AI Risk Management Framework (AI RMF). Second, a lifecycle approach instead assigns safety and security activities across the model development lifecycle, as in DevSecOps and the OECD AI lifecycle framework. Third, a threat-based approach identifies tactics, techniques, and procedures (TTPs) used by malicious actors, as in the MITRE ATT&CK and MITRE ATLAS databases. We recommend that frontier AI developers and policymakers begin by adopting the functional approach, given the existence of the NIST AI RMF and other supplementary guides, but also establish a detailed frontier AI lifecycle model and threat-based TTP databases for future use.

Motivation & Objective

• Address the growing complexity and evolving threat landscape of frontier AI development through systematic risk management.
• Identify gaps in current AI risk management practices by analyzing existing cybersecurity and AI frameworks.
• Propose a multi-layered defense-in-depth strategy tailored to the unique risks of frontier AI systems.
• Guide frontier AI developers and policymakers in selecting and implementing appropriate risk management frameworks.
• Establish a foundation for future development of specialized AI lifecycle models and threat intelligence databases.

Proposed method

• Adopt the functional approach by mapping essential risk management activities to core functions, as defined in the NIST Cybersecurity Framework and NIST AI RMF.
• Implement a lifecycle approach by integrating safety and security practices across all stages of frontier AI model development, inspired by DevSecOps and OECD AI lifecycle frameworks.
• Apply a threat-based approach by identifying and cataloging tactics, techniques, and procedures (TTPs) used by malicious actors, drawing from MITRE ATT&CK and MITRE ATLAS databases.
• Synthesize insights from multiple frameworks to identify overlapping and complementary risk management domains.
• Prioritize the functional approach for immediate adoption due to its maturity and alignment with existing standards.
• Advocate for the creation of specialized frontier AI lifecycle models and threat TTP databases to support long-term risk mitigation.

Experimental results

Research questions

• RQ1 How can existing cybersecurity and AI risk management frameworks be adapted to address the unique challenges of frontier AI systems?
• RQ2 What are the key gaps in current risk management practices when applied to frontier AI development?
• RQ3 How do functional, lifecycle, and threat-based approaches to risk management complement one another in securing frontier AI?
• RQ4 What role should standardized frameworks like NIST CSF and AI RMF play in guiding frontier AI risk management today?
• RQ5 What foundational steps are needed to build future-ready threat intelligence and lifecycle models for frontier AI?

Key findings

• The functional approach, based on the NIST AI RMF, provides a mature and immediately actionable foundation for managing frontier AI risks.
• The lifecycle approach enables systematic integration of security and safety practices across all phases of frontier AI development, enhancing predictability and control.
• The threat-based approach, using TTPs from MITRE ATT&CK and ATLAS, allows for proactive identification and mitigation of adversarial behaviors targeting frontier AI systems.
• A combination of all three approaches—functional, lifecycle, and threat-based—creates a comprehensive defense-in-depth strategy for frontier AI.
• There is a critical need to develop dedicated frontier AI lifecycle models and threat TTP databases to support future risk management at scale.
• The paper establishes a clear roadmap: adopt the functional approach now, while building the infrastructure for lifecycle and threat-based systems in the future.