FEDERAL CYBERSECURITY RESEARCH AND DEVELOPMENT STRATEGIC PLAN (2019) tools to standardize privacy requirements from data owners and custodians. Such design tools should support end-to-end composable mechanisms that protect the privacy of sensitive inputs at rest and in use, resulting in outputs that are tagged for privacy policy enforcement by their consumers. • Develop efficient, robust, and agile privacy controls for distributed analytic applications that can be tailored to the privacy requirements and the available resources of each party (e.g., secure, multiparty computations). Techniques are also needed to automatically invoke and verify privacy controls in accordance with policies. Devise methods that allow composition of privacy rules of an aggregate or function output from the privacy requirements of each record or function input. • Foster techniques and models that can systematically assess and quantify privacy risks, such as the effectiveness of privacy protections and changes in privacy risk when disclosed personal information is correlated with other information. Develop metrics and measures of reidentification risk for disclosed and obfuscated data, both at the time of disclosure and obfuscation and over subsequent accesses, uses, and potential correlations with other data. • Develop models, techniques, and evaluation metrics for redress and recovery from privacy violations, such as provable assurance that personal information is removed from a dataset or is otherwise rendered inaccessible. Research techniques to identify and redress inaccurate or unauthorized personal information informing machine learning models. Secure Hardware and Software Computing today is geospatially dispersed—ranging from mobile edge devices to data centers—and highly heterogeneous in nature—ranging in capabilities from tiny microcontrollers, mainstream CPU, GPU (graphics processing unit), and reconfigurable processors to highly customized NPUs (neural processing units). Computing systems execute highly diverse tasks ranging from sensing, perception, planning, and data processing to complex scientific applications where computation and data may be offloaded from one system to another continuously for performance, load-balancing, and fault tolerance. Data and storage are geo-replicated, and the network evolves with time. Vulnerabilities and defects of any remotely accessible software can have significant cybersecurity consequences for the system. Software is developed with vulnerability and defect rates that force dependency on a detectand-patch paradigm that is expensive and ineffective. Against this backdrop, securing data, codes, and computation from hardware and software to network and storage poses significant challenges. The tools and capabilities of adversaries improve with time. Assumptions about threats need to be reevaluated and defenses need to be adapted constantly. The foundation of security is authentication of hardware, attestation of low-defect software, and secure software updates to migrate a system forward by revoking compromised assets and patching newly discovered vulnerabilities. Secure root-of-trust provides such a foundation for these functions by establishing an unforgeable identity that remains inseparable from the computing assets and provides a small trusted computing base with hardware-protected cryptographic keys used for these purposes. Secure hardware and software are a required foundation for the modernizing of U.S. military forces, as identified in the 2018 National Defense Strategy of the United States of America.22 This foundation enables the development of resilient and survivable cyber capabilities, federated networks and information ecosystems, missile defenses, nuclear forces, autonomous systems, and other critical functions 22 An unclassified summary of this document is available at https://dod.defense.gov/Portals/1/Documents/pubs/2018-National-Defense-Strategy-Summary.pdf. – 19 –