The Defense Advanced Research Projects Agency has issued a broad agency announcement seeking technical proposals for a new program that aims to develop automated approaches to conducting cybersecurity assessments on computer networks.
The Cyber Agents for Security Testing and Learning Environments program focuses on developing an artificial intelligence toolkit designed to create realistic environments and train AI agents in mitigating advanced persistent cyberthreats, DARPA said Monday.
CASTLE will use reinforcement learning to train defensive AI agents in adversarial environments that replicate actual networks and simulate defensive actions of cyber threat actors against counter-APT tools.
The program will run for four years and will be divided into three technical areas that seek to automate the creation of realistic network environments, learn cyber defensive operations for maintaining operations and enumerate potential attack paths.
“Reinforcement learning may enable the creation and training of cyber agents that are much more effective than current manual approaches for addressing APTs in networks,” said Tejas Patel, CASTLE program manager in DARPA’s Information Innovation Office.
