|
Abstract:
|
Machine learning (ML) is being implemented as a solution to scalable defensive and offensive capabilities in cyber-security, ranging from semi-automated decision-support tools to fully automated capabilities. However, ML models can be exploited in at least four ways: 1) attackers can poison training data used to train ML algorithms to degrade prediction quality or redirect predictions altogether, 2) attackers can evade by manipulating runtime data to ensure ML models mis-classify malicious behavior as benign, 3) attackers can infer information about training data, and 4) attackers can approximately reconstruct the ML model for further analysis and exploitation. The DoD is committed to “leading in military ethics and AI safety” as one of five key actions outlined in the strategic approach that guides its efforts to accelerate the adoption of AI systems and "will invest in the research and development of AI systems that are resilient, robust, reliable, and secure" - https://media.defense.gov/2019/Feb/12/2002088963/-1/-1/1/SUMMARY-OF-DOD-AI-STRATEGY.PDF
|
