The CASTLE project aims to formulate network hardening as a reinforcement learning problem. Our novel approach to creating a cybersecurity AI gym leverages an abstract message bus that enables red and blue agents to interact with simulated and emulated cyber ranges. Agents compete in these cyber ranges to impact or maintain network workflows. Our hybrid simulation / emulation approach enables training at simulation speeds with emulation based realism.

As part of the 5G Deployment Evaluation and Performance Optimization Testbed (DEPOT) project, A2 Labs is building a versatile 5G Testbed which will result in a performance evaluation and resource requirements of various publicly available 5G solutions (OpenAirInterface, srsRAN, Open5GS, Aether, etc.), provide tools for metric collection within the testbed and offer deployment optimizations for any specified user service requirements.

As the need for the use of VPN services has dramatically increased, especially with the recent pervasiveness of remote work, the Software-Enhanced Networthiness Analysis of Commercial VPNs using Path Analysis (SENTRY) project will employ node probing and link analysis techniques to determine a path taken by a VPN tunnel, assess its trustworthiness and ensure that malicious intermediate nodes are avoided in establishing a VPN connection.

The RobustNet project aims to reduce log and packet capture storage requirements while still maintaining the ability to do root cause analysis months after an incident. Our approach leverages outlier models to identify repetitive data than can be down-sampled to reduce storage requirements. Our approach uses NLP and Sequence model based template extractors to measure similarity at both a structural and contextual level. But identifying redundant / repetitive data, we can achieve significant storage reductions beyond classical compression techniques.