DoD Applies Zero Trust to AI Model Integrity
The DoD is already applying existing Zero Trust protocols to its AI Testing & Evaluation processes. This includes using MLOps to ensure model integrity and mandating human-in-the-loop for critical decisions, extending ZT principles beyond networks and users to the AI development lifecycle itself.
The DoD's push for Zero Trust is driven by a strategy released in 2022, mandating all components achieve "target level" compliance by the end of fiscal year 2027. This framework is built upon seven pillars, including "Users," "Data," and "Automation and Orchestration," and details 91 specific capabilities that must be implemented. Leading this department-wide effort is the Zero Trust Portfolio Management Office (ZT PfMO), which orchestrates the adoption of these principles. The ZT PfMO recently approved the implementation plans for all 41 DoD components and is now exploring how to use AI and machine learning to automate and scale the continuous assessment and validation of these security controls. This initiative is part of a broader digital modernization effort championed by the Chief Digital and Artificial Intelligence Office (CDAO). The CDAO's AI and Data Acceleration (ADA) initiative, launched by Deputy Defense Secretary Kathleen Hicks, embeds technical experts within combatant commands to speed up the integration of data and AI capabilities. Applying Zero Trust to AI development directly impacts flagship programs like Project Maven, which uses machine learning to analyze intelligence data. Now managed by the National Geospatial-Intelligence Agency, Maven's end-to-end AI pipeline serves as a critical use case for securing data and models against adversarial attacks, from the cloud to the tactical edge. The DoD's methodology for securing AI models aligns with the NIST AI Risk Management Framework (AI RMF), released in January 2023. This framework provides a voluntary, cross-sector guide for managing risks throughout the AI lifecycle, focusing on governance, mapping potential risks, and measuring impacts to build trustworthy systems. For implementation, this strategy heavily relies on the "Visibility and Analytics" and "Automation and Orchestration" pillars. This involves using Security Information and Event Management (SIEM) and Security Orchestration, Automation, and Response (SOAR) platforms to get a holistic view. Solutions from partners like Splunk are utilized to ingest data across the seven pillars, automate policy enforcement, and provide the continuous monitoring required for Zero Trust.
