The National Security Agency (NSA) has unveiled a Cybersecurity Information Sheet (CSI) titled “Deploying AI Systems Securely: Best Practices for Deploying Secure and Resilient AI Systems,” providing comprehensive guidance for organizations stepping into the deployment of Artificial Intelligence (AI) systems. Released on April 15, the CSI serves as a roadmap for National Security System owners and Defense Industrial Base companies, especially those engaging with AI systems developed by external entities.
At its core, the CSI aims to elevate the security posture of AI systems by bolstering key aspects such as confidentiality, integrity, and availability while mitigating known cybersecurity vulnerabilities and fortifying defenses against potential malicious activities targeting these systems. The scope of the CSI encompasses a broad array of considerations pertinent to the deployment and operation of AI systems, with a particular focus on machine learning-based AI systems deployed either on-premises or in private clouds, especially in environments characterized by heightened threats. However, it does not extend to the utilization of AI deployed by third parties.
In terms of securing the deployment environment, organizations are advised to establish clear roles and responsibilities, foster collaboration between relevant teams, and ensure alignment with existing IT standards and policies. Robust architecture design entails establishing security boundaries, identifying blind spots, safeguarding data sources, and adhering to secure-by-design and zero-trust principles. Additionally, applying established IT security best practices, promptly patching vulnerabilities, encrypting sensitive data, and implementing strong authentication and access controls are essential steps in configuration hardening.
Furthermore, organizations are encouraged to adopt a proactive approach to network security by assuming a breach is inevitable, deploying effective detection and response capabilities, and integrating incident response procedures. Throughout the AI system lifecycle, rigorous validation processes, including integrity verification, version control monitoring, thorough testing, and supply chain security checks, are essential to ensure reliability and security. Exposed APIs should be adequately protected through authentication mechanisms, input validation, and sanitization practices to prevent unauthorized access and data breaches. Continuous monitoring of model behavior, facilitated through comprehensive logging, unauthorized change detection, and access attempt monitoring, is crucial for identifying and mitigating potential threats. Similarly, measures to safeguard model weights, including hardening interfaces, applying hardware-level protections, and isolating weight storage, are vital components of a robust security strategy.
In terms of operations and maintenance, strict access controls, leveraging role-based mechanisms, multi-factor authentication, and attribute-based access policies, are essential for minimizing the risk of unauthorized access. Effective security training programs are critical for enhancing user awareness and promoting adherence to security best practices. Regular audits and penetration testing, conducted by external experts, are indispensable for identifying and remedying security vulnerabilities. Robust logging and monitoring capabilities are essential for detecting and responding to abnormal activities and security incidents promptly. Staying vigilant and applying regular updates and patches to AI systems, evaluating new versions to ensure optimal security, is crucial. Planning for high availability and disaster recovery, including the implementation of immutable backups and secure deletion capabilities, is also essential for mitigating the impact of potential disruptions.
Keeping track of the everchanging AI landscape can be tough, especially if you have questions and concerns about how it will impact you. Don’t hesitate to reach out to BABL AI. Their Audit Experts are ready to provide valuable assistance.
